Join us for the bioXcelerate AI Launch Event, live from Inspace at the Institute for Design Informatics in Edinburgh. While the excitement unfolds onsite, we’ve set up this live stream to ensure that no one misses out on the incredible lineup of guest speakers and the ground-breaking discussions taking place today. Tune in and be part of the action from wherever you are, for an evening filled with insight and innovation.
Timing your timing is perfect well done especially for uh mostly academics I’m I’m impressed 2:00 and we’re all sitting down that’s amazing tell me where so everywhere around here I’m guessing I can sit here okay but I I can’t move a the left and right because you got it
Perfect I think that’s a good time to actually start with the event uh thanks a lot everyone for coming here today haven’t had the chance to speak to a lot of you but even a couple that I spoke you’ve come from different countries possibly even continents so that shows what a what a
Great event we’re going to have for you today and the reason for you coming here so that’s amazing so welcome officially to the bio accelerate AI launch uh we’re very we’re looking forward to showing you what we have for you today uh a couple of things though first I’m Harry
I’m going to be your MC for the first half of the day the second half is childcare duties for me so Chris will take over for that half uh a couple of housekeeping uh things there is no expected fire alarm to take place so if that happens we just go out from the
Door that we entered and we see ourselves across the road WC is right on the back and we have Jamie somewhere pretty close there Jamie which is our first dater for any medical emergencies Jam is right there uh we’ve got an exciting agenda for you today we’re going to start with
Some people from inside bio accelerate going to start with our CEO Alan cley then we’re going to continue with Chris and Zana two very important members of the by accelerate AI team and then we have three amazing external speakers for you uh Richard hio and Ricardo uh I’m going to keep it
Immediately to the important stuff you didn’t came here to hear me talking uh so I’ll start introducing Alan uh Alan is our CEO at bio accelerate AI with over 30 years of experience uh in customer transformation across marketing and intelligent service Consulting he’s worked with a lot of top firms Accenture
PWC sof Partners uh he holds an MSC in biostats and an MBA from this University and it’s this background in both science and business that drives his results oriented approach so please join me in welcoming Allan to the [Applause] stage the the first item was not to fall down
The stairs which I always didn’t succeed with but um apart from that so um it falls to me to do the briefest of prologues um for the day um uh and most importantly just to welcome everybody to what for us is a really special a special day is the culmination of a huge
Amount of of hard work um not only in a scientific and mathematical sense but also just the the business of pulling together and forging it and making it work and setting everybody in um I want to I want to say hello and thank you to some old friends some
Actually very old friends long-standing friends um who are in the room today there’s one or two very special um people who’ve come along today which I’m absolutely delighted about um a bunch of new friends um it’s great to and I’ll meet more of you um through the course
Of the day um and I’m looking forward to that um very important also to to acknowledge and recognize our academic and research partners and collaborators um because more of this later when um Chris and John and Cole and talk about our work um but that matters a great
Deal to us um that partnership with Academia and research and Industry um is really what Drive really what drives us so um great to see everybody here today um and then very little of it would be possible without our clients um and the the the the pharmaceutical companies and
Others we collaborate with um uh and we help pay our way um so never taken for granted never forgotten um and absolutely delighted that some of you are here today and um dialing in um and also others who’ve been interested enough to come along and hear what we’ve
Got to say so very much appreciated um and then finally um our incredible um bio accelerate uh team which is um a lot bigger than it was even a year ago um but still growing and still um looking to you know build um deter in a determined way over the coming over the
Coming months and years so it’s quite nice to have a little point of reflection um when you do these things um and if I think back two years ago the team consisted of one uh data scientist in the form of Chris Chris Foley um and Chris and I had a I
Was introduced to this um academic from Cambridge who was interested in doing something different um Chris and I hit it off pretty much immediately um and had a shared vision of where where we wanted to go uh with our data science and engineering Venture um and in particular we were really excited about
The prospect of being able to do what we are doing here today um which is to build a different kind of uh U business around life sciences and The Innovation data science and engineering Innovation uh mathematical innovation in life sciences so um um lovely to be able to acknowledge knowledge acknowledge that
Um we now have a team of 50 which is quite some going 2-e period um uh you know the gray hair is um age related but also activity related over the last 18 months um and our plan is to double that in the next uh in the next couple of
Years um so we’re on a a really big mission um bio accelerate is a critically important part of that um and Harry mentioned uh a little bit of a past in in biology and Science and so on um so it’s it’s I’m very much the lay
Person here today but it fills me with absolute Joy um to see how well the team has settled in the work that we’re doing the acknowledgement and so on it’s just absolutely fantastic um so just kind of briefly reflecting on you know what is it that drives us so I think the biggest
Thing is purpose um so we are all excited about the not the not just the the technical aspects of what we do but the purpose of it which is to improve life life chances life outcomes um uh uh in Health Sciences and that’s the biggest thing that you know that drives
That drives the team um The Innovation is is a function of of the power of data the availability of data the extraordinary scale of that um I think back you know 20 30 years when I or 30 maybe 40 years when I started out um it just wasn’t possible it just we
Didn’t have access to the kind of data that we do now and the we you know a lot of the mathematics was in place but we didn’t have the data we’ve now got the data um and that the Innovation centered around that is um is super exciting um
Uh I think it’s also really important to to kind of reflect on the value the importance of experts so we talk about you know being data scientists being mathematicians um being scientist um but the context is everything so the collaboration is all about working with leading experts in their in their field
Leading experts in Academia and also in our industry um industry Partners um and bio accelerates are newest Innovation um area so this is a big commitment for us um really really exciting um and I can’t be happier that than I am today to see everybody in the
Room and to share this with you um it would be remiss not to um also uh say thank you um to Richard um to hio um and to Ricardo for giving their time today um and uh um and helping us out with um no no doubt at all will be really
Interesting uh talks and and no doubt there’ll be some really interesting um discussion um following on following on from that so really that’s pretty much it for me um other than to say I hope you enjoy the day um uh there’s been a huge amount of work’s gone into this hats off to
Leanne and Jamie and the team um for everything that they’ve done um it takes a lot of effort to pull this together um so thank you lean and team for that um I will look forward to meeting everybody um later on catching up with a few
People haven’t seen in a long time um and uh enjoying the day along with the rest of you thank you very much [Applause] uh by the way I forgot to mention that one of the m i don’t usually do cards but one of the main reasons that I want
To do cards is the impressive background of all the people here and I was afraid that I will forget something if I don’t have a card uh so uh we’re going to continue with uh Chris and Zan uh our chief data scientist at bio accelerate and our director of Health Data Sciences
At by acceler two very important members of the team uh a bit of background for each one before I introduce both of them for a joint talk if I’m right so Chris uh Chris started with a mathematics degree from St Andrews completed a PhD here again at at edra focusing on
Creating machine learning tools to understand diseases like cardiovascular disease he also spent some time in Cambridge as a research fellow uh specializing conal inference and human genetic studies uh uh he’s notable for creating open- Source ml software and so and using AI to solve Newton’s equation of motion uh showcasing his Innovative
Contributions to the industry thank you for making us feel bad Chris that’s great continuing with uh Zana uh similar she holds a PhD in maths and stats with a specific focus on developing methodology for analyzing multiomic and Neuroscience data sets prior to her role at bio accelerate she spent four years
At a as a genetics data science lead overseeing the R&D development of proprietary software for genetic data analysis and population stratification and her work notably concentrated on neurodegenerative projects including Parkinson’s disease uh F Lana has also spent nearly two years uh has nearly two years of experience as a senior scientist in
Translational genomics where she led the integration of genetic evidence into Target identification and validation processes using causal inference tools so uh it would be great if we could actually welcome both Chris and [Applause] Zanna is the clicker working so the intention was not necessarily to have a double act
But given that we speak several times a day uh I’m sure we can pull something together sha and uh yeah while we wait for the slides I guess what I can add is obviously just to say a big thanks to Alan Allan is a pillar throughout the whole of the
Organization um I have a number of a handful of very important me mentors for me personally in terms of my career and Allan is definitely one of those and um it’s been quite the journey as Alan says to go from single digits as in you know
An integer one all the way up to 50 people across a a small number of years and I was obviously lucky enough um to meet janana who’s really been um an absolutely fantastic uh leader in the Health Sciences world as well uh right I can play on with a bit of clicking I
Believe we just do this clicking is not an option this is where I get my tutorial okay wonderful okay BR yes so as you say thank you thank you to everyone uh for coming as well um there’s a number of quite an array of people here actually
Uh you know people who have helped us with our sort of you know the the development of our licenses for our products so it’s wonderful to have um a couple of the legal team that helped us here um there’s scientific innovators there’s business innovators Etc so um
Very very pleased to see everyone uh I’m going to continue the discussion uh and be as quick as possible going a be in in some of the time to talk about improving Health outcomes through Cutting Edge scientific Advance advancement so any organized ation study technology Etc has
To be driven by an outcome and of course bio accelery are no different from that um and I thought sort of long and hard a little bit with seruan Janna how we we BR bring this this to life so we’re very much focused on health outcomes and
People um driven by data driven through industry and the infrastructure there so I thought maybe this this map from a paper in 2016 in the lanet might help us so this is a global map clearly and what we have on the left hand side and the colors toote this are
Different um diseases and conditions that are most prevalent across different countries and really what I want to bring to life here is there is no country that is exempt from having health issues health related issues and there’s variance uh within that as well um to bring to life the timeline with
Data and health outcomes and our really really are understanding again I’m going to turn to this 2016 paper in the lanet so what we have are uh age groups stratified by age bands roughly five years and we have two time points so we have 1990 and uh
2016 and what we’re really focusing on here is um years living with disability and what we can see is in 1990 there were a lot less years of people living with disability than in 2016 now there are confound factors here we’re getting much better at measuring data accumulating data understanding
Disease and the definition of disease is also a very sort of um slippery um uh exercise at times but you can see whichever way we cut the cut cut this you can see an enormous um burden particularly on society to really understand living with years living with
Disability 3.5 million people in the UK live with a rare disease 15 million live with a long-term health condition so even in this room we will know people we may well ourselves be living with with um particular Health outcomes that we would like to you know Advance our
Understanding of and also pull forward our ability to uh provide treatments and therapies and Therapeutics this inevitably takes us to the mechanisms with which we really drive forward the creation of new drugs and all roads lead to phac Pharma and biotech companies so we’ve got to think clearly about what are their challenges
And bio accelerator absolutely focused on this so from a drug Discovery perspective what’s what’s keeping people up in the farma industry at night and why is that important well we’re spending more money than ever before on trying to find new drugs but actually our ability to identify targets is decreasing over
Time the development cycle roughly takes 10 to 15 years so covid was an exception not the rule so the ability to find a therapy drive it through the treatment cycle the drug development cycle the clinical cycle normally takes roughly 10 to 15 years and this is the thing that we’re
Really really really focusing on is the vast majority of our decisions that we make early on are wrong and in fact one in 20 are correct in that they go through all phases of clinical development onto in into primary secondary and tertiary care so we want
To take that 5% and turn that into a much more healthy position and it turns out and this is why we’re here really is that targets initially with human genetic evidence are twice as likely to make it through that cycle now this is that that measure
Is in its infancy so we believe we can do a lot better than just twice of course we are no company at all without ambition um and our our firm ambition is to be a market leader and go to AI partner in in silico drug Target identification and validation so what do
I mean by in silicle I don’t often just like to drop in some Latin what I really mean is that these are datadriven targets um fundamentally a nice a nice sort of um statement you know we’ve worked quite hard to refine I think uh it was you know regardless regardless of
How divisive the figure was and Churchill said if you want me to talk for 2 minutes then write to me two weeks in advance if you want you want me to talk for 30 minutes write to me five days in advance if you want me to talk for an
Hour I’m ready and I’m yours so actually getting sentences like that and popping them into position is is quite a difficult one but fundamentally what we do is we use Advanced statistical and machine learning Technologies to discover new drugs that’s the bit that’s the that’s the bit that we’re focusing on what are
We famous for this sounds a little bit egotistical but the reality is that you cannot enter into this space without expertise Health requires lots of time and effort so as Alan said I’ve been a mathematician for 20 years and focusing on Health in particular for the last 10
Years and it takes a lot of time you know the math is one thing the contextual relevance of the application is more important far more important so what we’re looking at is developing novel mathematical Solutions to understand the genetic basis of disease to then therefore better understands the
Athology of disease and effectively help pinpoint causal mechanisms in the form of genes proteins and ultimately the interplay between multiple disease uh biological outcomes that lead to disease so we were building vast knowledge graphs not too dissimilar to how um Google sort of won the game in terms of
Search enges Etc so so what is our impact it’s all about accelerating time to Insight so data are there the vast you know pools lakes rivers oceans of data are there the ability to drive the insights out of that is an incredibly subtle and difficult mathematical computational biological Challenge and
That’s what we focus on so we’re accelerating the time to Insight to draw out of the the the kind of reservoirs of data that we have but our continued focus is now and will continue to be topologically driven by actionable biological Insight towards making sure that we minimize the
Risk of failure in clinical trials big thank you to Sur for putting together our Cube and our go-to um sort of sticker for Bio accelerate so what I’m going to do is pull the Hat up here so I’m going to focus a little bit on bio accelery so the reason why these
Three kind of elements I mean of course element of Academia industry is well founded so the reason why the three exist is because we want to sit in the interplay between these two domains so lots of innovation in Academia lots of innovation in Industry historically Academia has been very open
And transparent we we we love to publish the latest and greatest in our our thinking right we get lots of um ego points from that but also it’s trying to move the dial forward from a scientific perspective but the same Innovation exists in Industry the distinction there
Is that often for propriety reasons for um commercial ADV advancements Etc and there are other gdpr issues ETC we want to set that data out sorry set those R&D exercises outside of the public domain bio accelerat is here to straddle both of those so we’re very very focused
On open and transparent science that’s applicable to the latest and greatest challenges in Industry so that’s what we’re doing so how do we bring that to life well let me just pull that leg and arm uh um of the cube out to the left so
How do we bring that to life well we have offices in the University of Edinburgh we have wonderful collaborations with academic um um senior academics uh across multiple institutions in Edinburgh and and Beyond we have you know Professor Ricardo marioni who’s going to be speaking later he’s been a long-term collaborator for
Bio accelery we also work with Harvard mits the broad Institute University of Cambridge Imperial College London um the connection between the the bio Bank in Finland and the broad Institute is one that we’ve we’ve we’ve had quite a lot of success with um so really we’re positioning ourselves to understand
Where the Forefront lies in an academic context and then stitching that together over to the industry uh Partnerships so we work with um five Global Pharma and biotech companies who are all um processing and have implemented products that we have developed in their drug Discovery pipeline today just now as we
Speak um and of course it’s the the collection of these three dimensions that we’re looking to to position oursel very firmly I’ll leave you with this the way I view this and I think we collectively um can sort of again compact a compact sentence here we’re looking at this from
A 21st century business model in the health sector so that is open and transparent research for the scientific and and Industry communities to enjoy to peer review to make sure that we’re watertight but to deliver that with commercially positioned products so the advantage here and this might not be particularly well known is
That when you build and develop a novel algorithm say that sits outside of application there might be wonderfully appetizing and mouthwatering scientific mathematical insights but the ability to translate that into something that sits within a pipeline that practical implementation can require huge amounts of adjustment and REM maneuvering so if
You start with the outcome and drive the science along with that you’re minimizing the differential there and that’s what we’re doing so um we’ve published in some of the world’s top uh journals in nature along with some of the the world’s well biggest and most well-known bio sort of um Pharma
Companies and working and leveraging uh insights from uh biobank initiatives like the UK power bank which is a wonderful resource for the UK and for the world more broadly with this I will hand over to my esteemed colleague sh kin thank you everyone uh first of all I
Just want to highlight how important is what Chris just told and talked about which is our motivation and why we are here and the underlying reason why we’re doing all of that so now going to talk a bit more about how we are bringing that ambition to life or essentially how the magic
Happens um I’ll briefly explain that bio accelerate is part of the Advanced Data practice at Optima which as Alan and Chris said is now uh more than 50 highly skilled date scientists and Engineers that work all together to combine their expertise and deliver fantastic solutions for clients across five
Different industrial sectors of which of of which the pharmaceutical is one um in BIO accelerate what we’re working on mainly is addressing challenges in Translation or genomics Precision medicine and digit and digital Health uh and we’re doing that by leveraging all of that inter sector expertise to
Guarantee impact at scale we do that by working together with our colleagues at the machine learning and decisioning science division that help us to deploy some deep learning and uh computer vision Solutions and we’re also achieving that by working with our colleagues at the data and software engineering department um and that are
Making sure that actually our impact is also at a computational scale so now I’m going to try to put this framework together put some faces to it a lot of the people on this uh page are here with us today um and why all of this is possible why combining
All that Talent is possible is due to Chris’s idea when he joined the company and he planned to see that well optimal is doing a fantastic job in data science we can expand to health care as well so that um seat germinated down the company uh and a lot of people want to
Join our journey so if you’re interested in talking about our commercial and strategic um View and and plan definitely go and chat to serou or to Andrew kampbell uh we’re also joined by um some of our scientific advisers today so Ricardo is going to talk a little bit
Uh later uh but also Olivia which is one of our clinical development advisors flew all the way from France so I really recommend you go and you chat to him um if you’re interested in some of our products or day-to-day machine learning and data science capabilities I recommend shatting to Su mju Daniel
Michael and Peter that have been doing a fantastic job and do which all that’s possible um what’s very important is that we have built here we have built a community of peers and our expertise complements each other and this is how we guarantee that we deliver absolutely outstanding solutions for our clients
And that our products are of utmost quality so once we have all of this Talent together and all these exceptional skills you you kind of have to find a way to funnel that and and to realize it and to translate it to your clients so that they can get the most
Out of it and we discovered that um the way we have a very flexible way of working to achieve this goal um and essential to meet our client needs we discovered that we are very good at combining both service and product to make sure that our clients are happy and
That we’re delivering the best solutions to accelerate time to Insight and I’m going to tell a little bit more about what service and what product is so through service what what we mean through service is that we actually work shoulder to shoulder with our clients on
A daily basis and in this way we’re actually managing to populate the scientific innovation Gap from a product point of view we realize that we can actually bu designing products and delivering those to our clients we can populate the technology and insight at a computational scale Gap and through
Combining uh these two um approaches uh we are actually guaranteeing um real life impact so just to now you ask well we have been doing all this great work and and you’re talking about it and you’ve been doing it for a while why are you doing your launch now well it took us
Some time to realize that actually service is fantastic and it’s very impactful but Building Product guarantees that you can also propagate across the industry rather than one or two people or two teams in Pharma company being able to scale up a solution and run it for five hours
Rather than six months we can now try to drive the whole industry forward and and have real impact so building that product engaging all of the engineers and the product team um took us some time and we thought that when we complete that which we have
Completed this week uh and it’s a big achievement for us it would be a fantastic opportunity to also officially launch our brand which is now about our motivation and our ambition combined with our products um plog graph is um essentially our Flagship product and the main aim of
It is to try and combine all sorts of multiomics data from transcriptomics proteomics mutilation um disease phenotypes and try to pinpoint the caal RO R causal role of genetic variants um that’s driven by this momic data I’m not going to fall into details feel free to reach out to the team I’m
Sure they’d love to chat to you about it we also have a lot of information about it on our website but what’s more important is actually the impact of that work and that baph is used to accelerate time to actionable Insight in drug Discovery and it also enables us to Der
Risk the drug trials the drug clinical trials um process in drug discovery so it’s really helping us to achieve our bigger goal and ambition um what’s interesting about plra is that for it to work properly we had to develop some other products on the go and one of them is also switch step
Which was designed to address the challenge of both accuracy and speed of genetic fine mapping now for those of you that don’t know what genetic fine mapping is this is an approach to try and pinpoint the causal Gene that are actually associated with uh disease and outcomes and we have managed to reduce
The process of a large scale fine mapping from 15 days down to less than five hours we achiev that with more than 99% accuracy and we guarantee a computational convergence which already brings us one step closer to our bigger ambition we also launched another product which is called um imp map and
It again uh addresses challenges of accuracy and speed but this time in genetic uh imputation for those of you that don’t know genetic imputation is the process of populating gaps missing gaps in your data that might be driven by technological biases such as missingness or low quality of the data
We managed to reduce a process that can take up to hours down two seconds and we did that by achieving greater accuracy than approaches reaching on average more than 97% accuracy across all types of multiomic data sets what this means is that by now being able to have a full
Complete picture of the data we can enable a complete Downstream analysis and identifying the right targets so to wrap all this up and uh this fantastic products that we have worked with um our team with their complimentary skill sets um in U introducing best practices best engineering practices best scientific
Practices combining them in product and best spoke Services we do believe that we have managed to accelerate time to actionable Insight for human genetics and Drug Discovery and today together with you we want to vouch for our next big ambition which is trying to go beyond the two-fold
Improvement in this y success so we believe that by deploying at the mass level these Solutions we can actually get a greater Discovery rate and success rate for drugs entering clinical trials through that we believe that first we’re going to save billions of computational full-time employee hours in running those analysis we also
Believe that we’re going to increase Health Equity and Drug Discovery through tools like prograph that enable analysis not only in one ancestry but across different types of ancestries so you can be more inclusive in your analysis and account for diversity in populations and we also believe that through this
Approaches we are going to reduce the lab based scientific resources that are timely and costly um for the industry so Chris I’d love you to contribute to anything else that you believe I might have missed pro thank you very much [Applause] everyone well thank you Chris and Zanna
Uh that was also sorry from my from my perspective a lot of the faces that you saw on that slide that Zan showed with the collaborative team and the different skills and experiences a lot of these people are here today so feel free to spend some time later with them during
The networking event to find out more about them uh so uh next up uh we’re going to have Dr Richard Weise uh again a a bit of an intro and preparation before the slides up here there Jamie uh so Richard joined cure Parkinson’s in 2007 during the early
Stages of the Charities formation uh transitioned from a background in cardiology and genetics and he shifted his Focus towards Parkinson’s research uh recognizing the need for innovation in that area Richard spearheaded the development of innovative program known as the international linked clinical trials and his efforts Have Been instrumental in
Transforming the landscape of neurological research uh to date over 4,000 uh individuals with Parkinson’s have participated in this research with recent trials showing promising results and that includes those involving type 2 diabetes medications and drugs used in lver disease treatment and wait for it in recognition of his contributions
Richard was honored this January with a member of the order of the BR British Empire for his services to paran’s disease medicine so if we can all introduce welcome Richard right now to the [Applause] stage thank you for that well thank you for inviting me to speak today that’s um I hav’t seen
That that text for some time um look I’ve picked up on tieler want to cover today and I’ve interl them sure check something your is your’s your now it’s working perfect okay yeah um I want to pick up on on what theend is that um accelerator set for us I want
To give a view from the nonprofit sector because one of the latest speakers wants to give a view from the academic sector so um we we have a huge number of uses or for AI and for machine learning in what we do and I’m going to give
You it literally could have been a one week lecture because there are so many paret is incredibly complicated finding our way through it back in 2007 when I started with pure Parkinson’s neurologists who treat Parkinson’s patients were telling me why are you trying to work on a cure you’re only
Going to give patients false hope well it’s changed now and in few weeks uh we’re going to publish a paper where we’ve completely stopped disease progression at all and we we’re making inroads in quite a large proportion of patients so for that’s spectacular but there’s so much more work to do and Ai
And machine learning have got a huge part to play in it so um that would be forward no okay I I just want to explain who we are so we’re we funding charity we fund preclinical and Clinical Research into many disease modifying uh therapeutic approaches a large number of drugs that
We’re testing is it’s um we globally whether it’s Academia or it’s industry and in both the pharmaceutical RNA and regener regenerative medicine fields we co-fund clinical trials with various federal and state governments other charitable organizations and many pharmaceutical and biot technology companies in the sense that a lot of
Pharmaceutical companies have given us drug and PBO for the for the trials that we’ve done uh which has saved us tens of millions of of of pounds in drug costs every year we we write uh an update on how many I mean what sort of clinical trials are going on in in Parkinson’s
And this was from about a year ago we’re just uh assembling the 2024 update uh as as we speak and one of the one of the uh graphs that we show diagrams that we show in in these papers each year how many uh and the nature of this so
Symptomatic therapies at the bottom I’m going try to move away where I don’t reverberate um so these are given to patients every day to make their day better but they don’t stop patients getting any worse year by year so uh the symptomatic therapies are at the bottom in verse
Phases and the disease modifying treatments are at the top it’s fair to say that we’ve been involved with about 40% of these in one way or another whether funding or or um some sort of involvement so our footprint in disease modifying therapies is is huge and we probably run it’s difficult to calculate
We probably one of the biggest um uh drug repurposing um initiative in the world so this was mentioned before we um we we run this International link clinical trials initiative and it’s been we’ve been doing it for 12 years now uh to evaluate prioritize and repurpose existing and new developing medications
That may have benefits in Parkinson’s so we’ve published a lot of papers about this and we have a huge key opinion leaders group these are the good and the great in Parkinson’s typically professors of Neurology who also have huge um Cutting Edge labs in various areas so we’ve got about five
More recruited since this slide was made but it’s a very big group we we have face Toof face meetings for two or three days every year typically in Michigan and we decide what drugs we’re going to put into trial next and this this is a typical vote every
One of these colors is a Committee Member vote for a particular drugs so we write these very very thick dossier about each of these drugs and we have tried to use AI to do that so I’ll come on to that in a second but they’re hugely curated um about the
Drug mode of action but also safety and um pharmacokinetics and much else besides so this is a process by which we decide which drugs to put into tral and this was done I don’t know three or four years ago um just took it at random quite a lot of those uh drugs are
Currently in our trials so we’ve we’ve become known for our uh gop1 trials we’ve done a lot of them um exenatide is used to treat diabetes and we’ve done I think there’s been about five or six trials now we’re currently just finishing our phase three trial of this compound but we’ve looked
At various other drug classes uh sorry various other drugs in the glp1 class including the agde which we um did a trial in Los Angeles andde we did that in 21 hospitals across France in patients soon after they were diagnosed we’ve done to trials that patients are much more severe further into their
Disease um and as we and this is in the same it isn’t quite the same drug class but it inhibits an enzyme which which impacts with that drug class um we’ve uh this trial is a multi-arm trial so we we have multi-arms against the placebo and all
Those drugs are over encapsulated so no one knows which drug who’s getting which drugs but that L gping we’re doing in Australia Australia Australian government gave us about $30 million to uh to set up a big fiveyear program there well this is about half or maybe a
Third of some of the trials you can see it’s busy I think that’s all I really uh really should say so the theme of the meeting that have asked us to address how Healthcare organizations as Pharma Charities NHS can utilize Ai and machine learning uh products to accelerate and
Increase the efficiency of drug Discovery and development so that patients can get efficacious treatment uh treatments more quickly and at lower cost so those those are bio accelerates words so um but when we think about it the perspectives of those different organizations differ greatly so you’re going to hear
Um from uh Academia a bit later on on this this topic but also the time frames out lines above span a decade or more so drug Discovery from lower cost when the healthy economists come in at the very end we we heard in a slide earlier it’s 10 to 15 years so
Um we think that Ai and machine learning have have got their places in all of those various stages of drug velopment and we’re involved in all those stages every single one of them the uh the drug Discovery and the chemical changes of structure of the drugs right the way
Through the clinical trials and the health economics we we in one way or another we deal with all of those and we think that Ai and machine learning is really going to uh make a huge difference so um I want to give some examples of how we a global charity work with around
60 pharmaceutical and biotechnology companies but we also work indirectly with hundreds of hospitals including the NHS um some some of our trials involv 80 or 90 hospitals around the world um so I want to give some examples and they’re they’re almost picked at random because
As I said I I could give a huge number of examples but I’ve just taken a few and and one thing I would say about Parkinson’s because I did work in uh an easier therapeutic area earli in my career and um it’s it is immensely
Complicated um and and so are the lot of the other neurology conditions and we only have uh rating scales for much of the clinical observations and I’ll come on to that in in a second and how AI can help but um we’ve had to find our way in
Lots of these are where no rout was was we had to imagine how to go forward I think Carl gust young said if the path ahead of you is clear you’re on somebody else’s path you know somebody else is is your boss and they’ve told you what your
Path is but if you’re creating it you’ve no idea you just find your way and we’ve had to do this in various areas of of medicine um and Drug development so want to pick up and is better more reliable diagnosis of Parkinson’s patients so this is more of
A hospital thing but it’s very important for our recruitment to clinical trials so it can it it happens that sometimes it can take two to three years to confirm a diagnosis of Parkinson disease since there are other conditions that look very similar in the early stages so I’m going to talk briefly
About machine learning approaches to improve clinical diagnosis so you can nail the diagnosis cut out those two or three years Ai and machine learning enhanced approaches for biomicro analysis that’s also if you get good biomarkers then then you can um help diagnose patients but you can do a lot
More because you can monitor patients in clinical trials and you also you can decide what patients should get what drugs so really taken it random because there’s a lot of machine learning and a um studies but this one is a machine learning method to process voice samples for identification of of Parkinson’s disease
So so what we’ve got here is uh normal healthy individuals on the left and Parkinson’s patients on the right now this was published in the paper and they’ picked a lovely clear diagram where to our eyes we can see there difference here and these These are fast for um trans transformed uh
Spectrographs of of voices but they can pick this out even when it’s quite obviously to our eyes we can see a difference but but the uh AI can pick it out even even when it’s it’s far more difficult to uh pick out which is the patient from the from the
Controls so we’ve been working for many years on biological samples uh that will be informative for Parkinson’s both to diagnose and to use as markers in clinical trials and to use as markers for whether the drugs we’re given them are working properly so this is very recently published Alpha slean
Is a is is a protein that we all need it’s used to um releas hormones and neurotransmitters it’s part of that release process and it does a few other things but it can form clumps and it can turn from what we call monomers single lines and they can form diers and tetramers um
And then it makes sell dysfunctional and that’s most Parkinson’s patients have got this problem not all of them but but a large the large majority so as we speak we’re trying to to redefine um Parkinson’s based on the biology so we take examples from um from spinal
Fluid um we’ll probably be able to do this um in blood samples um on a regular basis to come and this is this is extremely important for the field because once you can nail down a a biochemical marker that’s really informative and you know exactly in which patients it’s informative then
That breaks open a lot of how we design the trials and and how we choose to give which drugs which patients we’ve been doing some exosome analyses where in some of our trials these are these are released by the brain If you um if you think of membranes in the brain there’s
A gene called bin three that just nips off a little recycle it and we um finishes up in the spinal fluid and also in blood we analyze the contents of that because it’s actually contents of the brain that are inside those recycles so we’re using this as a as a technique to
Understand the results of some of our trials that’s worked very well I want to talk about some of the databases we have um at our disposal so the first one that was set up was the Parkinson’s Progressive markers initiative ppmi and and um and this is patients with Parkinson’s year by year give
Samples and have all sorts of clinical examinations and imaging and this is available for everybody it’s open source and so we can do some wonderful work on on this um and so far there’s been more than 2 papers scientific papers published on this data alone so you can see how informative it is
When when open source uh patient data like a great volume like this comes out so um this is gp2 which is related to ppmi gp2 think of it as hundreds of geneticists in 70 or 80 countries around the world all piecing together because there are Geographic differences in
In uh in the genetics that are related to Parkinson’s and one of the forms of uh Parkinson’s is aligned with a ash kazian Jewish um Heritage line another ones come from North Africa of course they spread uh another ones come from the Philippines uh so the uh gp2 program is hugely funded
So about 400 um uh PD geneticists are flown into a particular location the last one was in Copenhagen took over a huge Hotel um for best P of four or five days and um and they presented all their data to each other and and they’re about 40 subgroups
And they’ve been given tasks and a lot of it being written up um and so what is currently known is that there are a few main causative genes uh and most of those have got mostly U incomplete penetrant and Alpha inclean you can get mutations to that Gene but
You don’t always get the disease you proba that’s what incomplete penetrance means for the non- geneticist here but there are about another 150 known genes from all this work so people working on this now from around the world we got 150 genes that add to the risk of getting
Parkinson’s however by adding expression quantitative trait Loi analysis and additional transcriptomics to the classic gas techniques which find us those genes and applying Ai and machine learning approaches a large number of additional PD related genes have now been discovered not published but they’ve discovered um but uh colleague of
Mine Anthony Cooper in Sydney and Justin as Salan in Oakland published this not too long ago discovering genetic mechanisms underlying the uh co-occurrence of parkes and nonmotor traits and they’ve used these um uh the these techniques uh to add to what we can do from classic genetic analysis
And this is some of some of their work in that paper so I’m just signposting it for you I don’t have time to go through what is an very impressive piece of work so let’s move on to ASAP ASAP is related to ppmi and gp2 but this looks more at
The underlying biological mechanisms and it’s all funded under the same umbrella and I don’t know if I’m even allowed to say how many of dollars has been invested in this program but what I can say is that it’s all open access when as soon as the analyses are
Complete which is phenomenal and I’m working on a on a project which um with them that it will produce literally millions and millions of samples and as soon as it’s all assembled and analyzed I can’t say what it is um but uh that will all become public
Domain that’s that’s the idea and that’s and that’s how this uh huge funding exercise that is Asap uh gp2 and PMI is configured and I wish other therapeutic areas had this but they give us so much data that we can work on absolutely primed for use in Ai and machine learning
Approaches so here we can see ASAP collaborative research Network that’s basically about biological mechanisms uh gp2 and aggressive Parkinson’s Mark initiative so let’s move on to drug Discovery again just a little snapshot um but also drug modification so we were using forms of AI from 2013 right way through to 2021
I’m delighted to hear Olivia Del ruse here put your hand up Olivia so I haven’t seen this man for eight or nine years and we worked together we helped each other on on some projects that that uh is totally related to this and some other things that you were doing in
Other therapeutic areas but uh come and simulator would love to talk to you so um so our our own curation of our big dossier that that we present to our International committee led to discovery of lots of uh new disease modifying drug candidates um and for a decade that
Diligent curation just sitting down for a year and writing a support for the disease modifying potential of a particular drug because it’s got this biochemical property and it works like this in in Asma and that’s linked and you know just forming um the what AI should be doing
Much better than we can but but our experience was we wrote about 220 highly detailed dosses like this over a decade describing these candidates for our International committee and we’ve launched a huge number of Trials through that process but over that decade only 1% of those dosses came from AI
Approaches because um cation was doing better for us but that it’s clear that that proportion is dramatically going to change and AI will will um will um in this particular area um uh it will it will help us amazingly that we don’t have to uh write for months to R
Research to to get this done another issue is where Ai and machine learning is said set to help greatly is the modification of certain the structure of certain drugs to Grant the more appropriate physic physical chemistry or other properties that we need to be able
To put them into patient so for us at uh cure Parkinson’s fosol is a good example so we funded cure Parkinson’s funded a a drug screen we knew that this protein Paris which was suppressing the number of mitochondria in cells uh because you need mitochondria because proves energy uh it
Was suppressing that it was suppressing another very important biological process so when Paris went out um then these other important processes went down so they we could inhibit Paris then and Paris is an act very long acronym but it starts with I think Parkinson um we we wanted to find an inhibitor of
This so we uh funded uh valina and Ted Dawson at John Hopkins in um in Maryland to um to do a drug screen and we screened 230,000 drugs to find an inhibitor of Paris actually we found seven but six of them we couldn’t give to man because
They weren’t safe but the seventh we could it was called faros and I wrote this um this commentary with Ted and Gina talking about how we would get this into patients but then we realized that okay it’s a fantastic inhibitor but we did need to give patients 27 grams of a yeah
Not practical is it so we want we were looking at Nano formulations and we were looking at uh other ways to change the molecule slightly so that it got through the gut and got into the brain much better um and in the end we found that
We we couldn’t really do it I mean maybe some fantastic structural chemists could but so last year I found myself um Walking The Halls of MIT talking to um uh RNA uh Specialists who could do that uh for us by a completely different approach but AI could could have helped us get
The structures right so that we could get more of theug from inside the gut or we could have injected it same problem it wouldn’t get into the brain very well um we think that an AI approach would would have been if we knew how to do it
At that time would have been good but we’re now on an RNA track for this very important biological Target so let’s move on to understanding biological mechanisms in par we don’t understand the biological mechanisms we’re not going to deliver the right drugs to the right patients understand why why they’re working so we
Feel we need Ai and machine learning approaches to help us understand certain fundamental biological processes far better we can see where we’re failing to understand the the biochemistry of this and and we just need some help it’s uh it’s just um hard to work out how some of the
Drugs are working and what compartments they go to and how they affect this and and what What mechanisms come in to replace what they’re um there’s compensatory mechanisms cutting in and all it’s very very complicated so there’s the lack of clarity about poorly defined patient groups is a very good
Example of this how do we select which patients will clinically respond well to which drug so I mention into the the problem in neurology if you’re in cardiology and you you’re using a blood pressure drug you your patient’s got a blood pressure of 165 you give them a
Drug you get a hard number and it comes down to 131 so measurable what have we got in neurology and I’m talking also about Alzheimer’s multiple sclerosis um epilepsy um motor neuron disease we got racing scal and the neurologist sees the patient and he weats from only zero to four speech
Facial expression rigidity finger tapping hand movements toe tapping leg agility posture and the list actually goes way down here and that’s only one quarter of the patient related list and then we’ve got other racing scales about sleep and um cognition so racing scales is not a good way to be running medicine
So what do we do when we when we’ done a clinical trial and we’ve got two groups of patients here they’ve both been on the drug we’ve got placebos somewhere else on another slide but these have all been treated so this group of patients 65 to 70% of them have a
Clinically meaningful Improvement in other words those ring scales have changed in various you saw so many rating scales but they have changed enough for the patient to feel the benefit you might be able to measure it at a lower level but if the patient can feel it that’s a really important thing
So we got one group of patients with 65% 65% of of whom respond with a clinically meaningful Improvement we got another group only 30% respond and also we got 10 10% of clinically meaningful worsening in that group as well which we don’t have in that one so this is
Um so I went to other experts this drug is used in other therapeutic areas so I went to The Experts the word experts in in that therapeutic area and I said but we’ve got this group and you know what can you explain it to us I said well we
The same we was hoping you explain it to us so it’s just a a puzzle that we we we have to solve and I think the answer is in a detailed analysis of the data so we’re just next door to the B Center I walked past it in the rain about an hour
Ago how Bay going to help us well this this paper was published fairly recently application of Bas approaches in drug development starting with a virtual cycle so the structure of clinical trials in neurology is almost um almost abstracted the partly because of the racing scales but also if you’re seeing benefits you I
Might some of our trials go on for two years some for three because the the rate of decline is maybe let’s say 5% per year per patient so if you’re comparing that with a control it’s very difficult to separate it out after a year we can do it and we have done it
But two years is better um and the three years is is even better except that the patient start to change by the third year depending on on the severity at the beginning but Bay’s approaches if we introduce them to Parkinson’s and we can do this retrospectively over hundreds of
Clinical trials will allow us to shorten uh our timelines and allow us to bring in historical Control Data allow us to do a lot of things this was really helpful in the covid vaccine story and um so I’m I’m looking forward to improve our clinical trial design massively by
Going by you through the data of of hundreds of uh clinical trials some that we’ve done some that others have done and I I think some sort of machine learning approach here would U would allow us to shorten our trials so patients just meet the uh timeline uh for change and then they
They’ve passed that base test and so we con shorten the trials which makes them millions of pounds cheaper as well so I think I’m just coming on to the end so use of um of AI and machine learning in deciphering biological processes in Parkinson’s so understanding and manipulating the
Properties of drugs of new of clinical interest for slowing disease progression an example of widely differing clinical properties of drugs in the same drug class that exert very different responses in patients so I’m coming I’m using go1 as an example here so these go1 drugs are used to treat
Diabetes but some of them are also used to treat obesity um uh by creating a lot of weight loss so we funded this this study two or three years ago which showed with I don’t know 12,000 patients in the UK and in Hong Kong that patients can be prescribed five or six
Different Med medications for their diabetes but if they’re on a gop1 Agonist it’s a it mimics a gut hormone uh they are 50% less likely to get Parkinson’s that’s huge that’s a massive protection and if they’re on the other drug that I showed you earlier Alpin um they which inhibits an enzyme
Um that breaks down natural gop1 in the body then you’re 40% less likely to get Parkinson so that’s a big difference so I’m currently designing with some others in Canada and and Korea a clinical trial of of prodromal patients which are Parkinson’s patients that haven’t yet
Got the Tremor but they’ve got some of the Sleep disturbances they’ve got some of the guts issues and they all had L of sense of smell that that um tells you that they’re they’re 85% likely to be on the track to get parking I so I want to
Give the Dual P1 Agonist to those patients uh before and maybe I can prevent them ever from getting their motor symptoms you know Tremors and their stiffness um of movement so so also one drug class is semaglutide which um is widely used now not only for treating diabetes but for
For for weight loss and lots of other things as well and then there’s uh the Lily drug uh to zapoti very similar results 15 to 20 kilograms per year lost lost in someone who takes these drugs I don’t want my patients to take these drugs because they’re going to lose weight and
They’re thin to start with typically but there’s another reason exenatide called exendin four here and lide those structures are very similar only a tiny change of the structure of the natural gop1 drug that has receptors it’s released by the gut and these these drugs uh have been changed by just two
Amino acids I think to stop enzymatic breakdown so normally gop1 is released as a hormone within a minute or two it’s been completely broken down by the blood it does its job and then it’s destroyed these stay active in the blood for many hours so that’s why they can often your
Um for over 24hour period um so we um they get into the brain so we’ve got here blood and at the top brain and in the middle there’s this Blood brain barrier that stops things like viruses and large molecules bacteria stops them getting into the brain as a protective process zenin 4
And Lio and some of the others cross the brain because they’re straight line go1 Agonist these are also called go1 Agonist but they’ve got this tail um as part of their structure semiti here it’s been the success story of the of the decade really certainly financially um but it doesn’t look it
Doesn’t get into the brain at all it doesn’t get in it it has it affects the brain by certain neurons tanos sites which come from a area at the bottom of the brain which senses the nutrients and the hormones in blood it doesn’t actually get into the brain to protect
The neurons like like this so okay so the two at the left show no entry into the brain so what we can do is start to redesign some of these uh hormones um and I think we need AI do that um but we do have exenatide and
Exde which we can uh start to give to patients as soon as we’ve got regulatory approval okay so I think uh perhaps I should just stop there um there we go well all I wanted to say now is that I’ve shown you diversity of of areas that you know we don’t
Understand that M that we found ways to make inroads and we’re beginning to help patients and we think some of these treatments will get into um use by regul approval but even when we’ve done that we’ve only affected benefit benefited maybe 30 to 40% of all patients and we still don’t know whether
These drugs will last for years and years to be Euro protected so there’s so much more work to do and I’m really looking forward to working with B accelerate to deal with it thank [Applause] you thank you uh thank you Richard a very exciting talk and also to see some
Of the ways that I’m sorry AI Big Data mathematics is actually used to improve the management of people with Parkinson’s disease moving from that we’re going to move to another very interesting talk uh and our speaker is online hopefully Jamie will multitask on the back and we’ll see hio from I think
Is is in Germany uh the topic the topic that hio uh will speak to us about is leveraging population scale biomedical data to accelerate the discovery of Novel drugs uh in a brief intro so uh hior run is the speaker formerly head of genetic epidemiology and currently medical
Director at Biogen uh specialized in translating genetic findings into drug Discovery and clinical trial design uh hio approach involved leveraging human data obtained through large biobanks disease Registries and real world sources to strengthen therapeutic hypothesis and ensure that clinical programs were conducted in the most informative patient subgroups uh as
A board certified uh medical geneticist and translational scientist High co-p possessive extensive experience in all aspects of clinical genetics with foundational trainings in Pediatrics uh finally prior to his industry tenure in 2014 Hao served as an assistant professor and independent group leader at The Institute of human genetics and in the molecular medicine
Partnership unit it will be really nice if we welcome Hao online he’s not here with us it would be nice to do that hello everyone this is can you confirm that you can hear me hello can someone confirm that you can hear me yes we can fantastic yeah so
Uh I think the Optima colleagues have said that they would be able to drive the slides yes fantastic okay great um so maybe you can just show the slides can you see them hi now I can see them yes okay great um oh it seems like when
I speak up the slides disappear for some reason um don’t know how I changed um is it possible maybe for you to open your own PowerPoint slides and we’ll always be at the same page with you you can tell us when to change the the page
Yes we we can we can try to do that so um I’m now moving to the the second slide so it’s the um uh I guess well well just as an introduction thank you very much again Christiana and team for inviting me it’s a real privilege to to
Speak at your um launch event um we have had a a great collaboration extremely fruitful very energetic and engaging um for for several projects in the past when I was still working at Biogen as their their head of human genetics I understand Ben sun is also on the call
Who has been closely uh collaborating uh with us and it’s really an amazing Story how far you have brought now bio accelerate and I wish you really the best of success to replicate these type of um uh uh activities also with other partners because there’s a clear need uh
For that in the pharmaceutical industry and that’s what you hopefully see on this slide uh which is that creating really transformation on your trugs is a really difficult uh thing to do because more than 90% of the trucks that we are actually bringing into humans um fail clinical fils this is primarily because
Uh we have um issues with efficacy and safety to often predict that the drug is actually doing its job often depends on animal data and um information that is not directly replicating in um in humans when we invite them into clinical trials and in addition there’s some some safety
Issues that are often just not easily predictable in the cell or animal models that we use before we start clinical trials so um don’t know if you can do next click but you should see that essentially we are picking the wrong targets then you can move to the next
Slide um which is the uh drug targets with human genetic support have higher successes we’ve already heard by this uh from Chris um I don’t have the option to switch on my camera my apologies so this was not given to me as part of this webinar I just got this information so I
Unfortunately have to continue apparently blind it it’s fine you’re not missing much you would just see me um but I I hope you can see the slides and if you click probably three times you should see the full slide depending on um how we Define uh genetic evidence uh we can
Eventually increase this probability of success depending on how confident we are that uh we are really having a a clear link between a variation or several variations within a gene and the disease of interest that we are trying to um address with a a drug and essentially how close we can come to
Build a drug that closely simulates um a genetic link between um a variation in the genome and uh a risk for disease and that this is um genetics is is there to to stay for drug um Discovery is U really evidenced by that over the last
Couple of years we um have really a large percentage actually the majority about 2third of FDA approvals that are with uh drug targets that are support by genetics next slide please so uh there are several sources where we can get to genetic data um Richard has really nicely um uh introduced Parkinson’s and
One way uh to to access genetic data is through disease specific cohorts and here clearly Parkinson is one of the front runner diseases overall we also leverage genetic data by uh genotyping and sequencing individuals who participate in our clinical trials but more more and more we are utilizing data
As it is being ascertained in very large scale and U population bi our banks uh the main topic that I want to speak about today so these are very large scale and very well-powered um uh cohorts that often capture health and biomedical information on several hundreds of thousands of individuals
Often have um Health Care um or register data that are capturing um really a lot of of Health interactions of individuals um from cradle to grave um they have some clear downsides when you compare them to disease specific hords for example they capture fairly noisy real
World data they have um not that the depth of information for distinct disease so um Richard said that it would be difficult to to grade the progression of a um of Parkinson for example with uh scaled systems but often we don’t even have the information that would be
Required to really come close to using that sort of markers um that are being used in clinical trials in those biobank resources often we do not have at all um specific longitudinal measurements but nevertheless these data sets are extremely powerful and this is um just by the scale of those hundreds of
Thousands of um individuals who participate next slide please because what allows us um this these numbers allow us to do with these cohorts as we conduct um what we call All by all Jas uh where we can link genetic information that we obtain in those cohorts to the very comprehensive phenotypes that are
Not just limited to one single disease but maybe Rel thousands of Health Data points um in those individuals and um very recently we have um added to to this component really also omic data sets where large scale intermediate markers like um a transcript form or proteome uh parameters are being
Profiled from diverse tissues that are um that enable us really to Overlay this with genetic and the clinical outcome information and what we do with those data because genetics for us as bias geneis is really a fairly Solid Ground is that we conduct Phenom wide Association studies so we use the
Genetic variation to simulate the impact of um a drug on a um a distinct gene or Gene product and we conduct Association study between variance in this Gene and the outcome measures of interest to us so for example a phenotype one where um we if we find an association that might
Protect from that phenotype we may get a support for efficacy for a drug or there’s of course other phenotypes because we’re not selecting only one but really multiple ones we get support that a drug that we would be designing against the distinct Target is also effic for other indications or just the iners
Um because we may find associations in opposite direction instead of a protective effect we have a disease um risk increasing effect there might be suggestions or indications for Adverse Events and all this information really can be utilized to predict the efficacy and the safety of drug targets really at a very early
Start and the next slide please just goes a bit into the details of what we do um in human genetics um to to to Really refine our information and come closer to a drug I think you can click twice and you will see the full slide or
Three times you can see the full slide and I may not want to go into detail here a lot um it’s bit difficult without seeing the slide um but just on the right side that’s a couple of uh activities that we um uh where we have been collaborating with uh Chris and
Team um to come up with new tools how we can really improve our um confidence that a gene is indeed related to a certain Gene function and uh that the it is indeed the causal Gene and the directionality of our drug hypothesis lines up with a genetic
Effect next slide please slide seven um what I do want to spend a bit of time on is now the exact vignettes so the two databases that we have been using most and you can immediately click again um all of you probably have very well aware
Of UK power bank which is probably the largest uh repository of its kind where more than 500,000 individuals have been recruited during the years 2006 2010 and meanwhile we have sometimes close to 30 years of um longitudinal follow-up information of those Biogen was part of Consortium where we have generated um a
Very detailed sequencing data whole exom sequencing data on nearly all of those uh participants these data are freely available and we have generated resources that allow us now to link those genetic data to um very comprehensive set of phenotype information and this data is all public there’s other consortia and initiatives
That are really constantly maturing the the wealth of um information that is accessible uh through UK B Bank next slide similar to UK B Bank we have been among the drivers of a finen which is a um resource that has now reached about a similar size as UK biobank this is now a
Research effort that is come into place by about 13 farmer companies teaming up helping to generate such resources is extremely um uh expensive um typically these activities are happen all in the pre-competitive space so these different funders are really working together and in this case with nine National bio
Banks across Finland in an initiative that is supported by um the um university of hinky and coordinated through the Finnish Institute of molecular medicine primarily arop palotti and Mark Deli and um is is really aiming to um uh establish a similarly Cutting Edge Resource as UK
Power bank in a uh in a way that allows really to very securely access individual uh level data and uh utilize these information where genetics is linked to um a detailed individual scale Healthcare um information uh to uh the scientific interests of um a really large spectrum of users next Slide the
Reason why we uh moved to Finland is uh that uh fins not only um have um a very uh very very good health care system with um digitalized health records that uh longitudinally capture really large amount of healthcare information but also because uh they distinguish themselves through unique uh genetics
Where um Finnish population has uh undergone several bottlenecks in their evolutionary history or settlement history of Finland which is Led that certain genetic variants in fins um are um showing up much more frequently than in non-finnish individuals and this has also LED uh to the um opportunity in a
Way to conduct um genetic Association testing on such variants um and find a significant uh new genetic associations at much lower sample sizes than we would have to um do do or or generate data on um in if you would go to to other populations um so I don’t know where you
Are currently at the slide but you can really click click towards the end here we have uh the ability to link really this now very detailed genetic data to the very detailed Healthcare data in these Phenom white Association studies and if we go on to slide
10 uh then uh what I and and you can also click twice here what I would want to um briefly uh tell you about is the collaboration that we had with uh Christiana and team uh where we conducted a meta analysis so a joint analysis of UK B bank and fent so both
Resources that I just talked about where we conducted an association testing in more than 650,000 um individuals particularly for coding variants that means those um variants that impact the protein coding um parts of the human genome um and we we try to Overlay close to 800 disease end points between uh
Those two data sets that’s shown on the right side that in principle these data uh points and disease end points match quite well between those two resources but I really have to tell you if you’re asking for a utilization of of AI in healthcare making matches between phenotypes across such type of resources
Is really um um a huge uh opportunity and I’m very sure with large language processing models this is where we will uh be able to much outscale what we’re currently doing in our case it was really two MDS sitting together and trying to make uh these these matches
Fairly manually but in general we have a a very large and well curated set of diseases we have a large number of individuals we have relatively variants um that allowed us to conduct a thorough Association study and this is shown on slide 11 I don’t want to go into any
Details for that study but uh what we did find is really many um novel genetic variants particular now at a u minor frequency of less than um two in 100 individuals that means really at the fairly low end of the um genetic variant Spectrum at a at an area where before um
These type of of um data sets have become avitable there was no association testing uh possible and this is particularly important because it now leads us directly to the most likely causal genes and instead of having very um to do a lot of laborious Downstream analysis can typically be quite
Confident that um the the the gene that we are hitting with those associations is really the one that we then should also follow on for our drug Discovery efforts next Slide the part that I would like to highlight light where our bio accelerate colleagues have been
Particular involved in is that uh we we were asking well we’re now going into those th those fins with the assumption that we indeed need much lower numbers uh for um uh increasing Association power is this actually true and can you make um us predictions that our bet on
Finland as an investment area really has paid off and this is in fact what uh beautifully came out in a and was validated in a simulation effort that we did together with um bio accelerate where we could indeed see that um the um enrichment of um genetic variance in
This Finnish population has been a massive driver of um the association signals and for I think if you look on average we probably would have needed for many of those signals about 10 times the size of the population um in in in in UK B bank if we um for for coming up
With similar Association signals and what was also really important to find was that uh the rarer uh variant and the more enriched a variant was in fins the stronger was indeed the association power that means the return on our investment and this is really a strong argument that future
Such Endeavors should indeed focus on a populations where variant are enriched similar like in Finland and this brings me to slide 13 where I um wanted to highlight that these meta analysis as we are doing have been doing between finen and UK B are going to become really more
And more important um as has already been demonstrated in the G area where people were looking at a single trade and this just slide just shows an example where uh we conducted our colleagues conducted um um an association testing for schizophrenia where they combined European uh data sets with data sets
From East Asians uh so two two larger scale Jas that just by putting together uh two different theocracy uh ge individuals from two different geographies have led to a massive increase in the associated law sign with this um uh the the the the starting um material for us to find new
Uh drug targets and if you click once you should see that one of our main questions is so if we see that sort of examples what are actually the cohorts that uh um farma Partners should best invest to gain a maximal scientific insight and uh because there are so many
Opportunities in principle some um tools are really helpful or or could be helpful to make and help us prioritize next slide and this is exactly where again we uh teamed up with chrisen Team uh where uh we we thought of can we actually use the available data to
Entirely reinvent how we conduct gen white Association studies that um in the past have really required um a very thorough curation of individual level data um and fairly laborious um analysis of and and and quality control of those data which had to happen at an individual level and then required an
Enormous compute time to um to conduct results and and and gain results even for a single trade let alone thousands at once and next click and this is where um the idea of in silico Jas came up where in principle we felt like well instead of going for many individuals
All one by one we could just um utilize um the the bio accelerate tools to Define an uh default or average individual that would allow us to really um then um predict uh the outcome of um uh of of Geno associations by just utilizing four parameters essentially information on the overall uh cohort
Size with a number of cases the number of controls as well as the frequency of the variants that you would want to study in cases and controls and this is now really a really exciting tool as as you can see from slide 15 the next one
Where as as one feature we can utilize a deep FR frog sampling to extrapolate variant Association cells in hypothetical cohorts so instead of really um doing this laborious effort going to to find uh people who are willing to donate a blood sample as well as their biomedical data um and and then
Conduct very laborious um and costly genotyping and um um Association testing we can essentially just um make predictions of well if we would do this in this cohort what would actually be the outcome and um indeed we just hypothetically generated virtually uh enlarged cohorts and um um and tested if
We would use this isg’s um uh approach to predict what would be the outcomes how good would that be and as you see in this graph on the left is that we really successfully applied this for example at the case of schizophrenia again where we compared um where we would have ended up
If a um a Jas that we did in 2014 um would have been extrapolated by is gas to the gas that was now coming out in 20 2022 and you could see that just this in silico approach from bio accelerate would have predicted 75% of the new Association signals that
Left coming up and only about a quarter of them were either Subs significant or incorrectly predicted so over all I think this is still a click tools like this isgs may indeed guide investment decisions into future um biobank projects and with this I’m at the end of
My presentation um thank you very much for your attendance and for for for listening and I hope it somehow worked out also with the slides thanks very much perfect perfect perfect uh thank you thank you hio and yes I can confirm it worked perfectly sorry for making it a bit
Extra difficult for you we had to test you uh finally for our last Talk of the day before uh we have a break uh you’ve heard and seen I think Ricardo’s name mention in different slides also we have Professor Ricardo marioni chair of molecular epidemiology of Aging at the University of uh
Edinburgh a very close partner collaborator of the bio accelerate team and he’s going to speak exactly about that uh the topic is partnering with bio accelerate to accelerate genomic research A View From Academia uh his passion lies in the exploration of Aging with a particular focus on unraveling the genetic and
Environmental factors contributing to health cognitive aging and dementia in recent years his research has focused on DNA methylation investigating its role in gene regulation and its interaction with genetics lifestyle environmental factors uh Ricardo and his group specialize in integrating multiple omx data sets including genetics epigenetics and proteomics to deepen our
Understanding of dimension’s molecular underpinings and refine risk pred prediction Ricardo all to you thank you so much Harry so today I’m going to give a very brief overview of kind of our perspective from Academia on working with bio accelerate and what advantages and benefits we can have from these
Partnerships so as Harry mentioned I’m professor of molecular epidemiology of Aging at Edinburgh uni so molecular meaning genetic proteomic epigenetic types of data epidemiology meaning large cohort studies like the biobanks we’ve already um heard about aging is a slightly more challenging term to Define um whether that’s purely as
Chronological aging and the passing of Time how does that intern relate to biological aging so the Aging in our cells organs and so on and then what about healthy aging how do we go about defining that is it absence of disease is it subjective well-being lots of
Questions about what is aging and how we should Define it so on the next slide I’m going to show you a fairly graphic illustration of Aging um I’ll let you decide whether it’s chronological aging healthy aging or um biological aging but please um keep your answers to yourselves
Um so this is um my profile through science so started as an undergrad in maths and stats at Aberdine uni came down to Edinburgh to do an msse in operational research then a PhD in genetic epidemiology followed that with a postdoc Fellowship in Cambridge in biostats and epidemiology um was lucky enough to
Spend some time in Bordeaux as part of that Fellowship then back to Cambridge then back to Edinburgh on a 10e track position which also gave me an opportunity to spend time in Peter viser Lab at the University of Queensland and um now back in ed as a 10e track
Professor so I think there’s um a few kind of pretty striking observations in terms of Aging here you can see that I’ve clearly lost my hair um lost my eyesight and um interestingly the only time I seemed to be able to smile was actually uh during PhD and postto times
When I was doing data analysis all the time um read into that what you will but in terms of Aging how we measure it and how we relate that to healthy aging and disease risk that’s really what motivates our group so in the cartoon here we have um very very
Simplified um overview of our work so we’re interested in collecting biological data bio samples whether that’s blood saliva urine um we also like to measure phenotypes whether that’s things about the here and now so lifestyle factors um or things in the past things in the future so disease risk through linkage brain
Imaging and then we try and use um statistical and machine learning approaches in the middle to identify patterns from the bio samples that associate with these disease outcomes and to break this down even more simply we have lots of data in terms of what we collect and in the
Middle we have methods I think this is a space where bio accelerates from the previous talks and from our experience of working with them they’re really thriving and creating some fantastic tools for the genetic epidemiology cohort and for um biom medicine so one of the cohorts that I
Work with and um we’ve worked with um bio accelerat on as well it’s called generation Scotland this is a study that began in 2006 we recruited around 25,000 individuals to the study from 5 and half thousand different families across Scotland so up to four generations per family interestingly we have um data
Linkage to health records on almost everyone in the cohort we’ve genotyped over 80% of the population and about the same number with DNA methylation so back in 2006 when the um cohort began we invited individuals to attend a Baseline Clinic where there were hundreds of questions so kind of
Pen and paper questionnaires there were also um clinical measures were ascertained so things like cholesterol lipids um in terms of the questionnaires things about lifestyle so alcohol diet um feedback on pain other diseases um more medical tests as well so things like lung function and um also cognitive testing
Depression questionnaires all sorts of um data were collected in addition to that we had the blood samples that I previously mentioned um linkage I’ll go into in a bit more detail on the next slide but crucially for um the cohort as well we also um our volunteers very kindly
Agreed or the vast majority um agreed for us to potentially Recon them in the future so this could be things like Rec contact by genotype so if there’s a rare variant or common variant in the population and we want to identify individuals that carry that variant whether that’s for inclusion in clinical
Trials or to test out um or to follow up for whatever reason it might be we can um Rec contact those individuals and ask if they’d be willing to join studies so this is um studies that could be led by research by industry by Pharma uh GS is open to
Working with all partners for um biomedical research we could also look to rec contct individuals with specific disease that could be based on the self-reported data at the start of the study or it could be based on data linkage so here we have good linkage going back to
Around 1980 for a whole host of modalities So GP data Hospital data clinical biochemistry and lab tests prescription data so it’s easy to identify individuals who had diseases before they joined the study and those who went on to develop diseases after they joined and um donated samples bio
Samples for us to analyze uh the numbers on the right hand side of the slide here are quite out of date in that we have many million more records than this now available so from a data science perspective this includes um things like images as well and also a lot of um
Potential for um text Mining and um written records so we have lots of great data we have lots of great data scientists as well at the University so why collaborate with bio accelerate this is the main reason um maybe 30 years ago wingdings would have been a mildly amusing font
Um does anyone want to have a guess at what that says anyone fluent in wingdings no translation so I’m quite aware that um in Academia very often um when writing Grant applications when writing papers when communicating our science if we’re identifying potential biomarkers or from jwa studies we have genes or
Genetic variants that we think are of interesting could be clinically relevant yeah we might write these things in our introductions and our discussions and our grant applications but how realistic is that and I think only really by working with um industry partners with the likes of um bio
Accelerate um biotech and Pharma can we really start to drive questions where we can have a genuine chance at um translating findings into clinical practice so this is highlighted quite nicely in this um entertaining and somewhat provocative paper published last month um so switching around the
Phrase of all models are wrong but some are useful to all models are wrong and yours are useless um kind of a summary of some of the limitations that are um present in a lot of these kind of biomarker and um kind of drug Target studies that we might um produce in
Academia and really challenging to think how can we make these models more clinically relevant whether for the likes of the NHS or for drug Discovery within Pharma so some examples here of um ongoing collaborations between U my group at Edinburgh and Optima and bio accelerate so the individual shown here
Are Rob who was a postto in my group uh his background in biomedical science Dany who’s a background in Neuroscience and’s a PhD student Scott who’s uh an an optimal employee and um now a also a part-time PhD student on the University’s biomedical AI PhD program
And then on the right hand side Daniel McCartney who’s um a senior postto in the group and also a part-time employee of Bio accelerate and his background is genetics so quite a broad background from our group from bio acceler as well and we collaborated with Chris
Janana hio and Ben who have already been mentioned on a couple of recent pre-prints which hopefully will be out fairly soon as Publications using the proteomic data from UK biobank and this was to um for Rob’s paper it was to explore the um the genetic contributions to not just
Differences in protein levels but also the variability in protein levels and with Danny’s work it was asking do the um do protein levels at the time of the bio sampling in UK biobank do they associate with disease outcomes 5 10 15 years in the future so whilst um I think the work
There is really exciting um one of the limitations around dany’s project is that it’s perhaps a little bit naive to look at a blood sample at one point in time and then just a solitary disease outcome in the future 10 15 years in the future because this
Is or at least when we ran these analysis it was without considering what diseases had occurred prior to that time point and what diseases had occurred in between that time point and the the diagnosis of the focal disease so Scott did an absolutely incredible job in his msse project that
He did with us um last summer this was taking advantage of the data linkage in generation Scotland to the primary and secondary care records that we have from 1980 through to present day and here he was asking um or building um a visualization framework to look at
Diseases that co-occur and to look at trajectories and Pathways that occur prior to uh focal disease endpoint or flipping that round after being diagnosed with a disease what other diseases is one more likely to go on and develop so in the example here we have stroke on the right hand side as the
Terminal node and um one thing that’s perhaps a little obvious and we’re currently working to tweak is you can see that es schic heart disease osteoarthritis and a few other diseases frequently pop up in these Pathways so these are basically diseases that are very commonly diagnosed in older
Individuals so what we’re now trying to understand is whether these diseases or these transitions are occurring more frequently for those who go on to develop a stroke compared to the rest of the population so looking for enrichment of the um Pathways rather than just modeling um The observed numbers across
Each um transitions on the branches here and I think this could be a really useful tool when scaled up to biobanks so it’s nice that we’ve um piloted this in generation Scotland we’re now extending it to UK biobank and there we can ask of the enriched pathways are there any genetic differences across
Those paths we can use tools like in silico gwas that hio mentioned to run really quick screening analyses of these different Pathways analyses that would take quite a long time using um offthe shelf or offthe shelf methods and then we can ask are there differences in polygenic score by the different
Trajectories can and by integrating things like the prescription data does this tell us anything about who’s more likely to resp respond to drugs and so on and so forth so kind of approaching the end but just to say that generation Scotland while we have 25,000 individuals in the
Cohort at the moment we’re currently recruiting new volunteers so from age 12 and above anyone that’s living in Scotland regardless of where you were born is very welcome to join it’s a quick online questionnaire we then send you a spit kit in the post so instead of
Um individuals dro into a clinic for hours of tests this is now a very quick process and convenient so spit in a tube post that back to us we extract the DNA for genotyping and um methylation analyses we also ask um if people are kind enough to agree to letting us link
To their medical health records and also for the potential to be reconed so again trying to build up this base of individuals that could be useful for future studies for whether um academic or industry based so in terms of the next steps as as far as I see them from my um
Viewpoint in Academia bio accelery is developing Cutting Edge tools for data science of course we want to be able to access and to be kind of first users of these tools to apply them to our Innovative data sets and also to work with bio acceler and their other partners
So Pharma biotech and to do this to really kind of co-develop health research projects that have a genuine chance of um having an impact in terms of translation whether that’s at the um healthc care setting or at the drug development setting and with that just to thank
Everybody in my group um thank you to all the generation Scotland volunteers and research team and thank you all for [Applause] listening thank you uh thank you Ricardo and a big thank you to all the speakers for the day if we can so we’ve reached the end of the
First half of the event we now have H 20 minutes where we can grab Refreshments stretch our legs and for those live we’ll see you back in 20 minutes e e e e e e e e e e e e e e e e e e e e recording in progress e e
E e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e
I’m going to a random crap Festival it’s great as well for the likes of the biomed students and okay hello okay bro hi everyone we’re just going to get back together reconvene for the Q&A oh need spe hi Shana wanty to take a seat and recard Richard prize you away from a very
Interesting conversation no doubt okay while Richards is um popping his mic on and uh settling down onto the couch uh we have two new um members um today’s operation in the form of Paul agapa and mju daki um and both are experts in data um Paul has a deep and Rich expertise in
Computational biology um has worked with major Pharma runs um a very very popular and a collective in London of which he has been very generous inviting some of the bio accelerate staff to present uh mju is also a wonderful data expert he is an innovator at bio accelery and has
Driven forwards many of our products so Paul I don’t know if you’ve got a couple of words just to introduce yourself um and then we’ll do man well I was an immunologist computers happen to me like to do with many and ever since then I feel I’ve
Been the the person in the middle make sure that we get the informatics and the computation and everything that we need in order to understand biology and that that context that comes from biology is provided towards the informaticians and the number end of things uh as said at GSK at astroica and
Many many years is perhaps unfortunately at the University including bual college so thank you thanks Paul muuu um like him I started Life as a biologist U medical microbiologist at the University of Edinburgh and I’ve always been interested in computer science so I’ve um gone over to machine learning and AI specializing in
Evolutionary algorithms doing optimization doing evolutionary algorithm um and I do have experience in engineering as well so I kind of in BIO accelerate team I combine um Computer Science Biology um and operations as well BR we also have Ben Benjamin son online who is an incredibly accomplish accomplished scientist he’s a medical
Doctor and PhD in statistical genetics epidemiology also and has a solid track record uh then I’m not sure if we have and hio is also online then I’m not sure if we have oh hello hio we can see you amazing yes apologies for the technical issues before so very nice to meet
Everyone bro Ben do you want to do yes I think you’ve done most of the introductions there yes I sort of started off from as Life as a medic uh and then end up sort of venturing into sort of the research side using you know when bar Banks and large scale gamic
Approaches were just beginning to take off along with other multiomic proes like you know proteomic and metomic so worked in that field and then sort of moved also to sort of on the into industry uh where sort of translating the signs sort of into more a industry
Framework uh and sort of have been straddling between sort of the three pillars uh ever since uh and yeah and it’s can see there’s you know opportunities and challenges that uh you know this initiative would likely uh benefit yeah for thank you so much ben I Ben is joining us from Boston
Unfortunately the the time difference did not favor him to do a day trip uh and and hio is currently in Germany and heidleberg so thank you very much for both both for for diing in okay so I’m going to get the ball rolling and my apologies for the fact simile on the um
The agenda so so basically the theme which Richard mentioned in his presentation is how Healthcare organizations Farmers CH Pharma charity um NHS can utilize Ai and ml products to accelerate and increase the efficiency of drug Discovery and development so that patients can get efficacious treatments more quickly and at lower
Cost okay so that’s the theme that we’re going to sort of resonate around but I’m going to start with asking about clinical outcomes so data has absolutely transformed how we approach science so in terms of the volume of data that are available and the ability to capture dynamics of a physical system hidden
Hidden in there and no more so in the last 5 to 10 years so this is pretty fresh and my understanding is that you know we still need solid use cases to prove out some of the the Technologies and their efficacy so I’m going to start with what clinical outcomes might we
Consider to First prove and develop uh the value of AI derived insights so those that Target most prevalent diseases improving quality years Paul please yeah you can see I’m begging to reply to this uh so let me subvert that question a little bit to start with and it’s just
It’s to emphasize that in fact it is a very sensible question and in fact it is the core one that we need to answer here what are the clinical outcomes in any of this that we do uh because historically of course we’ have gone in medicine from the traditional what we call called
Modern medicine where we have a set of established practices that we believe actually benefit the patient so forth to the Advent of evidence-based medicine which at first seems ridiculous and absurd but you know the very fact that you know what’s the evidence for any of our treatments actually providing a
Benefit to the patient and that approach is actually sometimes given us some very uncomfortable findings at time you know that if we treat people for fevers antibiotics and so forth that actually they will tend to their illness will last four hours less say or that if that even there are
Treatments we found that when we use them that the outcomes were worse so we do definitely need to start talking about this what is what is the bottom line what’s the outcome here okay the patient the second half of that is why it’s also very necessary for these
Approaches because we live in an Age of Wonders where all of our Ai and large language model pixie does solve all our problems immediately and we’re moving into Nirvana where everything is done and none of us will need to know skills or anything like that but the vast majority of these systems
Fail when they try to put them into the actual Clinic you can look there’s a number of papers there there’s about AI models during the pandemic there’s uh series of papers by Laura winus and so forth that showed that if you got all of these models that were published in the
Literature the vast overwhelming majority of them were useless they could not be applied for actual patient outcomes now having set that up I’ll just very briefly talk about what is that we should measure and I think we should very much look at patient cetc measures in there there are things that
Are easy for to measure but they’re not necessarily the things that patients are interested in things they’re good for for example if you’re looking at cancer therapies the goal standard is overall Survival how long we live except this of course is it’s fairly brutal thing to measure sometimes very difficult to
Measure and often for a patient the thing that actually matters is how long am I going to be disease free what’s my quality of going to be so we need to measure things like time to next therapy or how long do you have of the cancer
Carousel before you go on to your next treatment and I’ll stop there thanks very much but that’s very insightful I think we often always you know we think in terms of the disease end point and reducing risk and maybe not some practical elements within that as well I
Wonder if anyone else would like to to jump in Richard you you look like maybe you have I mean we understand your neuro um neurogen focus in Parkinson’s but um yeah clinical outcomes so I’m really envious of uh people working in oncology uh obviously um alive and dead
Is um pretty clear as an outcome but also you can measure the size of a of a of a growing tumor it’s one millimeter it’s two it’s three it’s four um and you’ve you’ve got some very very hard measures and I’m I’m terribly jealous of that you know I’ve complained about the
Um about the outcomes measures we use that are are based on scales of one to five an opinion of a neurologist that does it depend on whether they had a bad day yesterday you know I mean it it’s really it’s something the um medical history books and we’re trying very hard
To change that I think this is a really important point and one that maybe is difficult to translate into the public sphere which is we think that AI can ingest all types of information as Paul has alluded to and suddenly invert I will do a little small dance in
Order to find a spot of low resonance um so that the AI will invert some of the difficulties in terms of Designing the data and it simply won’t it ingests those difficulties and maybe reinforces them and I think one of the demystification we often talk about the demystification of the algorithms but
Actually it’s about the measurement process in clinical outcomes as well and I think that’s news to a lot um that for instance in cancer and in cardiovascular disease areas we have quantitative measures that we measure through medical devices Etc whereas there there can be very informed but nevertheless opinion based measures that
We use in neurodegenerative disease come in so I think just to um double down on that point where in terms of neuro if you look at um the various subtypes of dementia and how you can see with the gold standard pathology how some um postmortem brains will be you heavily loaded with pathology
Yet the individual would have had no kind of clinical manifestation of the disease yet others so there’s kind of this compensation thing that some people are better able to um cope with the um the disease pathology but I wonder if we can maybe use um biomarkers omix and AI to come up
With datadriven um characterizations and subgroups of disease so you can identify those that are more likely to be or might or might be um less likely to be affected and are there um whether it’s polygenic scores or other characteristics that define subgroups that may respond better or worse to um different
Treatments that takes us quite nicely on to the next question but I will offer the opportunity for our esteemed online guests uh to maybe give their input there given that they’re both medical doctors as well I I I guess my addition to this is mostly sometimes is about the
Model that’s been trained and models that’s available to the data that’s been fitted so obviously some diseases the more prevalent ones are more likely you know to benefit from these data hungry models than something that’s rare where you just don’t have the data there and obviously that goes you know down the
Stream of what data should we be collecting and AI machine learning you know algorithm can also guide that in terms of these should be additional data which should be collecting to reinforce uh certain uh disease areas that are low prevalence I guess the other part is really is not about sort of necessarily
Sort of replacing uh you know the the clinical uh outcomes are here or you know how how a clinician does their sort of clinical workup it’s more at the moment is more about sort of using it as a guide as a co-pilot you know to suggest and so you’re not missing the
Important things as a reminder uh and also the modality I guess is the other things you know you know Imaging itself is probably more mature in the AI space uh compared to some of the um other sort of AI models uh and obviously the different modalities have progressed at
Different rates uh and so sometimes you know you can have you know radiological endoscopy live sort of uh AIDs that are driven by AI to help you for example pick up cancerous regions on you know slides the pathology slides or Radiology or you know in endoscopy uh setting uh
Whilst you know others you know like medical text at the moment is still a challenge because there’s a lot of in individual variation of how the text is inputed there’s a lot of Errors there’s a lot of noise there’s a lot of habits that different conditions take on so
There’s still a lot of challenges there that still need to be worked out despite the emergence of these language models so I hear there Ben is that there is lots of um proof of Concepts that are there to augment already established processes but we’re still working
Towards that great big leap that we can achieve um although there may be some sizable um steps that we’ve taken today BR hio I can I can bring you into to another question um uh if you if you would prefer so um the the nice transition from uh from
Ricardo there was to think about geographical considerations now that’s something that um hio had in his talk Richard had in his talk as well um and Ricardo obviously was speaking a lot about generation Scotland and the nuances of the Scottish population as well so in that vein how might we
Consider geographical dis differences in disease prevalence and characteristics such presentation in order to prove value and achieving fairness so one thing that often is talked about with machine driven learning is about you know reinforcing bias in data sets reinforcing already overly represented um demographics in
Population so so what can we do there in terms of geographical um differences I wonder who wants to put the hand up there what we’re doing is I mentioned it briefly earlier but we’ve now got over 200 genes which influence the um the likelihood of getting the
Disease and in our gp2 program we’ve we’ve got uh 400 PB geneticists working in 80 countries because the the genetics do differ from country to Country and I don’t know about well I’m pretty sure that’s true of a lots of other diseases as well but I think Parkinson has bitten
The bullet and just we recently reported three weeks ago I think a new Gene in um Parkinson’s related Gene in Africa you know and these are popping up all the time now in um in Japan in China you know the you don’t get that Gene but this this Gene gives gives you Parkinson
You know with some sort of autal recessive component or whatever so we’re learning all the time but we cannot do this in a Caucasian population and think we’re representing the world can I bring in hio here hio I wonder if we just do a mic
Check um yeah so I I I think um you you mentioned the biases that come from the different Geographic um regions but I think that’s already one of those uh opportunities yet to that you do find systematic ways to actually work out those biases using AI or um I I
Immediately think of um well phenotype imputation what I also referred to in my my talk where we can just use um hopefully well maybe existing or improved large language models to work out uh where do we have missingness of data how can we eventually infer um phenotypes or intermediate end points
From typically sparse data and um so that if we if we can really um apply that approach to these very large cohorts we would already um again quite a bit and could do a more refined Association testing than to the relatively crude endpoints that we are
Currently um applying to and if we would be able to do this really not cohort by cohort in a very much handy manner but systematically across um many of those I think this would very much accelerate um our our our way to towards conclusions
So I I I see it in a way as a as an opportunity currently and yes uh in terms of um um eal or or equal access and uh the opportunities to to Really deploy uh Tools in different geographies or ethnicities I mean that’s that’s something we as geneticists are working
Through since very long where we have our entire field focused on Europeans and this field is only catching up now um very very actively in fact in non-european cohorts after it not only has been recognized well this is underserved um um ethnicities where we have some obligation to go but also
Because we see wow there’s enormous um discoveries to be made and enormous um opportunities for um actually utilizing those um insights absolutely um yeah every challenge is an opportunity and opens a new door isn’t it and I think um you know this is something that’s known in The Wider um large language model
Community and maybe I bring in a little bit of Mandu MW here M’s time here um first of all the normality of the data state that the the large language models are trained on give us our first step forward but of course they’re not necessarily automatically appropriate
For the challenge at hand and therefore we have mechanisms to repurpose and repoint so if we have gold standard scenarios where we have better phenotypic um labeling or understanding then we can use um architectures which not to introduce too many um technical terms but retrieval augmented generation
In the large language model space that allow us to repurpose and man I don’t know if you if you want to to say something about that sort of pre-trained versus better using gold standard data to to to funnel that towards more appropriate what I’d like to add this
Discussion is about deployment of these um pre-trained versus your uh classic tools or your t b SP um tools deployment is going to be quite difficult to do without bias because of gdpr and residency data residency rules because for example some countries in the world have um this rule that says their data
Cannot leave their jurisdiction so if you’re are going to be doing a study that spans multiple countries you need to take those cons those issues into consideration so whether you’re deploying an llm or or B B pooke tool you need to take that into consideration and Design Your solution your study or
Your study to match those conditions I really like that and it’s Shameless plug um not to hios uh you know mention of in silico G which does exactly that right by summarizing and effectively anonymizing the data per region you can still aggregate as if you had access to each of the individual
Level data and I that is you know that’s really about the design The Challenge and the ambition to meet that challenge um shall I move on so once a successful Ai and ml driven approach to drug Discovery has been identified what are the challenges in scaling these Solutions so Ben you mentioned that
There are some exercises and scenarios where we’ve we’ve we’ve had success so how do we scale these Solutions and I guess um I can I can I can I can add a little bit more meat on the B and talk about things about reducing time uh and costs
Associated with bringing new drugs to the market uh again all links to what of the success is today so scalable Solutions is is is the key Shana can I can I introduce you to that one yes um of course so one of the challenges I see is access
To data so scaling Solutions means you want to do it either across different types of data sets and settings or you want to do it more and bigger and faster um so if you want to do it more and bigger and faster obviously depending on what institution you are in
And what type of access you have to data that might be very easy for you or it might be very challenging as smaller companies and biotechs and even charity organizations don’t have um access or they don’t have um access to large computational resources they don’t have access to Engineers to
Set up this overly complicated not overly complicated but complicated Cloud architectures that enable this type of scaling up um so um what Richard said about Open Access Data is incredibly important because um it will benefit um the rest of the the the world Beyond just large organizations and in solons to be able
To try and add value about scaling of this AI Solutions across data sets and across different um in not Industries but settings uh we’re entering in what you also said before that about biasing AI uh data is not collected in the same way we don’t have the same data
Protocols about collecting data even with the same diseases what Richard did didn’t say about Parkinson is that those ranges and those scales they also different between countries sometimes or between academic institutions some of them or the way they Define them um and and that’s true for many other uh
Diseases which uh inure the generation and also uh in in cognitive uh cognitive diseases or even talking about mental health issues so not being able to to have the same scalable comparable data uh and the same data collection protocols across different settings countries data sets um so that you can
Make um so that you can introduce those to the AI models without any bias um is just going to lead to a lot of garbage in garbage out situations I think that’s incredibly powerful um that actually it’s about data harmonization harmonization of protocols and in fact in clinical research and with the um
Principled application of statistics and data science we often look first at harmonization and standardization of protocols so even to the biobanks the way in which we’re enrolling participants classifying um disease status and common across um different centers Etc I think absolutely so first what I hear is first we should think
Really carefully about the data designning protocol um bro Paul as someone who sat at the heart of innovation [Laughter] I wonder if you’ve got a couple of words on this as well I I do a little I feel like I could almost summarize It Up by saying it’s hard
Uh uh I think this comes back to the whole you know we have a universe of proof of Concepts out there you know what is it that we can actually walk out instead of distribute and there are very much these real fears I think that AI in
A sense is a rich man’s game you know who could ever afford the computation to put together something like chat gbt and llama and all of those sort of things and so forth so there are a couple of different angles I think on this number
One is that and I mean I’ll boost you guys if you won’t things like in silico gws I think are incredibly important you know how can we run computations in an effective efficient way without needing to bankrupt ourselves so the second thing I think is there definitely there’s is to Echo this about
The harmonization efforts for data now uh this used to be a really fraud problem I should say I once worked on a an observational trial where the white blood cell content was measured in three different ways across the countries who were looking at three entirely incompatible ways but nowadays
We have things like omo okay we have data standards that we can use to make these things compatible comparable and I think the next step is going to be building that suite and that Galaxy of tools that can speak those formats speak those languages and help
Us get data into that format it comes back to that old joke about AI or data science sometimes that uh you know when people get interested in the subject first they think that they’re going to need a big computer and next they think they’re going to need a bunch of
Brilliant data scientists and they finally conclude that what actually is data we need to sort out the data problem or everything else is mood completely colleagues in engineering um couldn’t agree more and I’ll speak on their behalf Richard please yeah I I want to respond to something Z said but
Also something you said so I tried to make sure that all the neurologists Parkinson’s neurologists that are in in the trials that i’ I’ve launched and there are a lot of them uh a lot of Trials a lot of neologist that they’ve all trained in
The UK or Harvard or you know in America somewhere um and that means that the delivery of the um the understanding of of of the racing scales however bad it is it is actually applied the same way but here’s me in 20120 or 2019 I’m in Beijing they got three million
Parkinson’s patients there overwhelmingly more than any other country and I’m planning to launch Five trials there with with one of the leading pedan neurologists in China I’m breaking the rule because of of having them all trained he was trained he worked for 11 years in um at
Harvard before he went back to China but he’s kind of running the show there and there’s a thousand neurologists in China and they handle the three million Parkinson’s patients and everybody is linked to each other by wech it’s an astonishing Network and I was we were seeing what we could
Do you know you need you need about four hours to measure all the waiting scales of a patient if you got them on a trial but could we shorten this could how could we harmonize that and it was the plans were starting to get Advanced and then this game of Parkinson’s well China
Scored an own goal Wuhan scored an own goal and we couldn’t start because of Co and we still haven’t started but this um leading pedologist um in China we’re meeting in Boston in six weeks time and we’re going to try to relaunch that but but I’m part of why I’m saying this is
That I think that works for every therapeutic area all patients are linked to all the neurologists in China via WeChat and you can assimilate information by that network if you know you need local Chinese people to facilitate it because you need to know the infrastructure but you you it’s it’s
A fantastic resource that that you might anyone in this room might want to consider Ian that that’s an incredible resource of free text in order to effectively derive insights through pre texts mining scraping and depending on the kind of um maximum sort of input allowable in terms of and what the
Chinese government will let you do and we’re not sure we haven’t put that to the test but how much can you get out of the country but I I would be very happy just to do it in China and that that’s fine you know absolutely with three but
I thought I should let everyone know that that it exists and someone’s trying to prosecute it I just want to pick up on Paul’s point about um efficiency I think that’s a very important thing because it efficiency also drives equitable across all the industry as well as
Academia because at the moment you don’t want the big Tech to be the only ones being able to run these large models and obviously in order to you know for everyone to be running it you essentially need something that’s highly efficient and this could be coming through you know computational U
Methodology type of things all coming from basic you know approximations from the ground up through mathematical means I think obviously there’s that disconnect at the moment because you have people people who work in sort of the the pure space where they do something and it works and then pop it
An archive and then no no one takes it over until someone s of spots it and then makes use of it uh down the line in an industry scale and obviously there’s the disconnect because things are done in a sort of Academia way where you have you know breakthroughs in methodology
Doesn’t necessarily lead to a computationally or industry grade tool that can be used so there’s always that disconnect but obviously you really need that expertise to be able to sit on both side and connect that up in order to make these efficiency gains and that’s I
Think things like B is a very good idea in terms of being able to bridge that Gap and having the expertise on both sides to be able to piece together what’s really needed and how to actually realize that potential and obviously I think that goes on to know you know to
Say that we the currently a lot of things are just brute forcing it with better and better computers and hope things get faster and faster with better faster storage and access but obviously this efficieny gains are going to be more and more important uh even with you
Know large language models people now go let’s go for a much smaller one TR in a much smaller model and give an approximately good enough answer at a fraction of the cost and the computation I think that’s going to be more and more important uh as the data gets larger and
Larger your prognostication powers are quite incredible Ben because the next question is about how important collaboration between Healthcare organizations technology companies and academic institutions is are in terms of advancing the use of these these types of tools so I think You’ set a fantastic basis so I’ll introduce one of our
Longsuffering academic uh colleagues in the form of Ricardo and to maybe to maybe pick that up yeah I think Ben kind of summarized it perfectly to be honest and as I tried to allude to in my presentation I think it’s really really important for us within Academia to be open and to
Actively engage with industry biotech data science companies like bio accelerate to to Really c-drive projects where instead of kind of the token sentence in a Grant application or a paper we’re actually designing something that sure it may take five years or 10 years if it’s a
Um a Pharma based or drug based um approach but something that could actually have a chance of delivering because ultimately we all have the same end goal and I think it’s just it’s fantastic that we have companies like bio accelerate in Edinburgh which is a massive data science Hub to begin with
You I mentioned the the doctoral training program in biomedical AI Precision medicine translational Neuroscience there’s lots of lots of areas of potential overlap and I think of more academics talking to um more industry and biotech is is definitely the way forward yeah I mean completely agree I think the ability to
Contextualize the research towards the kind of practical outcomes that might be of relevance in industry and in commercial applications is is so important but at the same time to still maintain a sense of Detachment that allows you to pursue avenues that you are interested in in the pure scientific
Instigation sort of sense and it’s not that there is any one side or other it’s a very very sort of um difficult area to disentangle because there’s Rich science going on all over the place that’s very pure in terms of its outcome to improve patients lives across all communities um I like that
M talking about scalability and collaboration we see sit in a perfect position because um scalability in terms of data you spoke about but scalability can be tackled from a systems point of view as well so um software modern software is aimed for the cloud they scale effortlessly so you can run them
At scale is not a problem but our use case doesn’t fit that scenario that type of scaling it doesn’t fit but what we bring is a bunch of Engineers and software developers and experts who have expertise in traditional scaling and now we are getting exposed to data science
And health data science workloads and we are learning how to scale them efficiently so that you can actually run these tools without going bankrupt or without having to own a huge H HPC so you can start small and then scale up and down as required using all the
Experience that’s available to us well I mean you know forgive me for you know making it attempt to quote Winston Churchill earlier and now turning to Shakespeare but it was ever thus right um I think the point here is that you know first and foremost we are human and
There are human level practical um issues and collaboration constraints Etc that surface so they need to be solved initially and sometimes that’s about resource burden so do we really want to be in a situation where the largest in terms of financial footprint organization s are the ones that can run
These types of sophisticated analysis and therefore have first access to the sophisticated results versus more of a diffuse approach and and I think you know that that’s why it’s really important to think as as Ben has said as you’ve said um rre and as as Ricardo has
Said as well to think about you know time to Insight what’s a practical so we often think of this tradeoff between if it takes me a 100 days to achieve 95 % accuracy versus 100 seconds to achieve 85% accuracy where are you going to place you know what are you going to
Invest in assuming that the cost the same okay so um driving forward with this and I really want to to to sort of bring in hio Paul and Janna here and thinking about and if I start with hio I’m thinking about how do we enable companies and organizations and Society
To become matchfit forgive my my um terrible terminology um um to deploy these types of tools and win trust um that we are improving life’s and effectively prognosis for patients well that’s a super hard question I would say I mean you see that um on the one hand tools that are easy
To use and clearly fit a niche like jpt are being taken up in an incredibly speed and people use it for their purposes that are fit with really being figured out on the Fly I mean that’s for for private use this is actually really amazing you let sort of the strings
Fairly lose people come up with really creative ideas how to utilize such tools and sometimes maybe I’m optimistic they will find uh great purposes that then also are sufficiently convincing on the go with 85% uh uh um uh there of doing their their optimal uh job of of
Convincing uh Brad ly that this is the way one should go forward um I’m a much more uh on a much more skeptical note I’ve the experience is from being an MD working in a in a fairly um structured uh clinical environment being in clinical development and that’s the
Space that is really very highly regulated and super conservative in terms of following right end points making sure that we are really living up to to to gold standards and uh fairly um conservative in taking things that are just not 100% uh where uh where we want
Them to be and where our patients expecting us um uh them to be and then really also in in general a certain mindset that is uh really skeptical to towards a hype and uh actually quite okay to well take a couple of more years uh to to work out a thorough end point
Instead of jumping on what seems obvious to an early researcher to jump on so I’m uh I think there’s a lot of um uphill battles to be fought uh and uh especially from from the company and the the medical side those uh parts that are the closest to regulators and the
Closest to patients and um here I think the field really needs to live up to these enormous expectations that are currently being put on it can agree more Paul well I’m going to largely Echo what hio said uh through a a large part of my
Career I was called a B matician I think because people didn’t know what else to call me and if you’ve lived in that mure you’ll understand the completely stigan quality of the software that you have to use uh I think at one point when I was
Imperial we we even we formed a rule of thumb that have some there was a paper that described a new piece of software and the paper had dazzling benchmarks and show that it worked perfectly and things like that there’s probably a less than 50% chance that it actually
Functioned you know it was it wouldn’t run on your systems you simply couldn’t reproduce the results they’d run it on some peculiar cohort and this is very much in line with uh the work I mentioned earlier on Laurel winance it’s terrible so uh my suggestion on one side is that we
Have to raise our state of that game in building software for biom medicine uh at the current moment it’s an amateurs game a lot of it is just people throwing things together you know that have never thought about software engineering or well-written program things like that and that’s going to require efforts from
Funders from journals and the universities all of that to take the far end of uh that question so on the other hand I think there has to be a lot of pressure and effort put into consortiums and pre-competitive efforts okay because it’s only through those that we can build things that we
Can put into common use across the industry that will raise all boats if you will no individual company or entity as going to do that alone they can’t okay we have to Short Circuit this I I mean I really like that obviously we have similar training so
There’s going to be a bit of an echo chamber here H Shana I’ll bring in a slightly different point of view on the topic loved your answers by the way um I think it’s also about building trust between the AI community and the geneticist biologists and clinicians because um sometimes we we people
Mathematicians statisticians forget that there is actual people staying on the other side of the line people that every single day and every moment of their life are potentially suffering struggling whose lives depend on it and um Jans had a fantastic example a while back when they trained the super high Ultra Modern AI
Neuron Network to predict some retino disease Aller clinicians refused to use the results because they said I’m not going to prescribe and and something to to a patient if you don’t tell me which features exactly of the retina is this AI using to to predict that they’re
Going to have the disease or not I’m not using it so they spent another year retraining it using machine learning models that had interpretable parameters so they could tell the clinicians which which is exactly they they’re using to predict whether the patient’s going to get the disease or not because otherwise
They’re going to be compliant if they get my Mis diagnosed and we should always think about we should maybe think less about our um rock curves and our accuracies and you know oh my God I’m so much better than this other state of the class method and think a little bit more
About how do I explain this to the biologist or to the geneticist or to the clinician so they can really use it to to make an actionable decision on their patients life I I think trust is like a key word in there just like so much
Agree with that because it runs like a golden thread through all this work you know I don’t I don’t worry about the bus that I use to get here because I have trust that the bus work I understand it and so forth but if you’re going to put
A black box in front of me and say this will completely forecast what’s going to happen with this cancer patient buddy you got some work to do and on that note so um at Cambridge so Ben started his PhD on he was already on the clinical program and did um a
Sment to do his PhD and I think the whole department were utterly shocked at his ability to start programming and deploy some very sophisticated uh data science tools so someone who straddles both clinical and data science B I wonder what you think about um winning trust I think interpretability is very
Important because from different fields from different perspective is an inherent prior been put through the training that’s been done I mean not saying that medical training is exactly state-ofthe-art because you know a lot of the things that you get taught in medical school I I would say are pretty
Iaic and out of date and sometimes even I query you know what’s the evidence behind it but a lot of the things just follow through and you sort get imprinted into a way of thinking and obviously in order to adapt into something new most people need you you
Can’t make too much of a leap you make you need to make a smaller enough leap such that is intuitive and interpretable so as you know for the for example the retinol AI model that’s been trained obviously Black Box would be too much of
A leap but if you take a step back and you know if you mask certain places and be able to say these are the features in the retinol image that the AI model is considering and these are different between cases and controls then you can build that sort of inter the bridge to
Interpretability from the clinicians because they look at it go oh okay that’s also the area we focus on when we’re look at retinal diseases or other you know parts of of a funders uh you know for for various other things and and and I think there’s also that hand
In hand where your sort of AI can be used to form knowledge and in terms of accumulation knowledge and building on that uh and vice versa so at some point you know there’s things that wasn’t understandable at the point and then you know AI introduces some models where
They picks up features that are not necessarily easily picked up by a human being or conition but that are useful or have utility and then that provides a training brand for people to look look into those areas to see what are the interpretability Gap that we can sort of
Fill from a knowledge perspective I think they go these things go you know hand in hand and also I said Paul said once something’s matured enough you just don’t questioning it you know I think same another field you could say is about self-driving cars right at the
Moment you know would you get in one and just let it drive you down the motorway or or would you take your wheels off you know off or self-driving uh vehicle probably not right away I don’t know I wouldn’t I’ll probably still have my hand sort of vaguely on a cereal not
Fall asleep but still is one of those things that over time once you know there’s enough of a capacity that everyone’s using and there’s no problems with it you sort of you adapt uh and so of become a climatized to it I guess similar things with the aviation industry you know technically are
Probably a lot of the planes are on autopilot most of the time but you feel more comfortable because as a captain and a pilot there not necessarily they’ve got their hands on a stick all the time but you just don’t know that but you assume you know there’s that
Level of comfort when someone’s at the wheel very much like that so I’ll finish with a question that’s I think um very relevant for everyone I’m going to start with for Ricardo because I know Ricardo’s put in huge amounts of effort along with many many others the generation Scotland and my question is
Do we see much of the Improvement in health outcomes and prognosis being derived from the principal development of biobanks and resources like generation Scotland UK biobank finen um biobank Estonia Japan um buback Japan Etc short answer yes I mean I I certainly think it might not be concrete
But we’re definitely getting there and I think another important thing to consider with these cohorts is kind of longitudinal sampling if we’re looking at um response to drugs or we’re looking at the evolution of signatures over time these are really critical parameters for us to consider but um yeah I would say I’m
Quite hopeful and I I think um has a huge amount of potential in the data that we’re generating I think also looking in terms of the um solutions that are genuinely population representative looking to Scandinavia where there’s these population wide so truly population wide Registries in Scotland there’s potential for things
Like neonatal blood spots which have been stored on everybody here since 1965 um so like over 3 million cards on absolutely everyone so it’s yeah yeah say more I’m obviously going to say more data is better but um yeah okay so just a small riff off of
That do we think the connectivity do do what is your was your opinion then on a very focused collection of data on phenotypes from a particular population like saying in Finland or in Scotland or elsewhere versus taking a more holistic or general approach or do you see that
It’s a bit of a divide and conquer that you stitch together all of these principal Frameworks that are designed for local populations good question I think it’s going to be a bit of both really there are some things that are going to generalize well I mean we see that with
Some of the the particularly the epigenetic predictors that we develop they translate well from Scottish populations to um people of Indian Malay and Chinese ancestry living in Singapore um genetic predictors don’t do as well across um different ancestry groups I think there’s always going to be kind of local things to consider
Whether that’s um pollution and environmental exposure in different regions different dietary and lifestyle um practices so I think it’s yeah there’ll be some things that generalize very well but others were they need to be tailored a bit more specifically Richard you want to come in there well I think so there are 10,000
Diseases if you add them all up I think Ai and machine learning has a huge amount to offer um and you focus on the bio Bank angle of this so what’s likely to happen I think most diseases that get around to being studied and I think that’ll
Probably only be a few hundred of them what’s going to happen there so I think I think the when you look over the 500 diseases over the next 10 years that will be really analyzed very closely from biobank perspectives and I think what you’ll find is um most of them benefit by incremental
Improvements and I think occasionally there would be light bulb moments which transforms that therapeutic area into into a a a new um situation but I think we will get multiple incremental increases which after all is how medicine as always there there’s a new technique there’s a new observation or whatever so I think
We’ll have those incremental uh improvements in a lot of diseases by this approach and every now and again you you’ll have a lightball moment and um and the the other thing that I think will happen if we’re smart enough to pick it up is that patients who somehow don’t respond as
Expected quite a lot of those and there’ll be different situations but quite a lot of those those will finish up redefined as a as a a new rare disease because they don’t fit into that pattern and they’ve been analyzed by the biank and AI approaches and they don’t
Really fit their response is different they’ve been too rare uh to pick up by individual clinicians but you add you start adding them up and it’ll be redefining some diseases and we’ll be looking for new treatments for those I’m I’m giving a a decade long look at this going forward
So picking up on Richard’s Point Ben hio how long until we find a biobank derived or we we we visit a biobank derived Nobel Prize for medicine I don’t know UK bank has deserved one I hope this would come every every year and and similar uh the these sort
Of approaches so I just wanted in general be uh I mean it’s it’s um I I fully appreciate the um the uh well be being being being cautious in principle uh I I think there are already quite some uh breakthroughs when we look at the the the the biobank landscape I mean
Ricardo is a very humble personality I think he’s understating a little bit that already just with an off offthe shelf proteomics analysis in this UK bank data he was able to demonstrate that you have in of course not yet replicated in in single cohort and
There’s a lot of work that needs to be done you can um exceed the predictability um with with such of the Shelf tool of a established biomarket that is being used in clinical practice since um since many years so I think it is really a matter of how can we um gain
Confidence that these type of analysis are really living up to to what we see now in exploratory research early observations which we are really excited about we do need to have a good good benchmarking exercises we do need to have cohorts where we can replicate analysis um this will take a couple of
Years this will require additional Investments that that will require researchers that are actually willing to do this type of implementation research which um will take probably a while to catch up but I think the data sets that we already and the results that we’re currently already working from uh um uh
Have have a very high potential to really massively transform how we are doing medicine and how we are uh approaching the um discovery of Novel drugs absolutely and I guess lastly is just the I think the main challenge is really being able to harmonize in the
View of you know data privacy and you know rarest form of geopolitical climate as well I think obviously individual level data confidentiality identifiability at all issues which makes me feel that you know the importance of being able to derive summary level results such that we can still arrive at a mathematically
Approximate or equivalent solution without having to go back to individual level data I think would be very important in plugging that gap which you know plugging or jumping between the walls that currently exist between different jurisdictions and different between different corts and between different data uh warehouses as well so
Inable to you know the summarize summarized method are very important in allowing for us to pull these things together in addition to be able to do meta analyses so no yeah absolutely agree and just for the avoidance of any doubt whatsoever uh Ricardo’s plus one to the Nobel ceremony has been already
Called by myself okay so on that note um can I ask Daniel suou and Peter to come down and just to say just want to um say thank you very much uh if you pop down um Michael I think you should probably present yourself as well that’s that’s all
Right so all of the products that we are developing um The Innovation and to Paul’s point about that principled engineering and MRE mentioned it as well that actually that makes sure that we can achieve the goal that we’ve set ourselves the very ambitious goal in BIO accelery is because of these people and
I’d also like to thank our very accomplished panel and speakers and those who are thousands of miles away and uh maybe you know in the hundreds hio uh several thousands uh but you know we can measure we can be generous with our tens of hundreds um yes so thank you
To everyone um for for for a wonderful event and um that’s us thank you very much all thanks everyone thanks thanks Ben