AI-driven multiomics in health

Dr Susan Mitchell speaks to Dr Chaitanya Erady about multiomics, discussing the key themes and questions of this intriguing topic.

Welcome back to Making science work for health, the PHG Foundation podcast that explains the most promising developments in science and their implications for healthcare. We discuss the underpinning science, the ambitions for improving population health and the impact it could have on patients, on society and on the people delivering your healthcare.
 
If you would like to find out more about what was discussed in this episode, you can find additional information on our website, https://www.phgfoundation.org/insight-and-analysis/ai-driven-multiomics-and-health/.

If you have any questions about the topic then you can email us at intelligence@phgfoundation.org.

Transcript

Copy Transcript Copied

Ofori: Welcome to Making science work for health. For this episode, Dr. Susan Mitchell, Director of External Affairs speaks to Dr. Chaitanya Erady, Policy Analyst in Biomedical Science about AI-driven multiomics. What is it and why is there such enthusiasm for its role in improving health? 

Susan: Hi, Chaitanya. Great to be talking to you today.

Chaitanya: Hi Susan. I'm really glad to be here as well. 

Susan: Yeah, so today we're gonna be talking about AI and multiomics, so I know we've done some work on this before, but should we just have a bit of a reminder about what multiomics is and also what AI brings to it? 

Chaitanya: Yes, definitely. That's a very important question as well, 'cause when you ask what multiomics is, there's like a simple definition that it just means combined analysis of multiple different ohms and that includes genomics, transcriptomics, proteomics, and the more, sort of, recent radiomics and pathomics as well. 

Susan: Brilliant. And then how does AI come into this? 

Chaitanya: So what we, what a lot of people recognize is because of the scale of this data, it's really passed any sort of manual analysis, and we do really need some really powerful computing resources, one of which is AI to try and integrate all these different types of data and analyze and sort of decipher meaningful biological signals from them. 

And that's why we call it like AI-driven multiomic analysis. So while multiomics is the crux of it, AI is sort of driving this moment forward by allowing us to infer meaningful signals from the big biological data sets.

Susan: Wow, that sounds really interesting. And it sounds like it could actually be many different things. It's not like one single thing, is it? We are looking at different types of ‘omics using AI and other ways to really analyze and sort of reach into that data in really new, exciting ways. 

Chaitanya: Yeah. 

Susan: So what have we done? Obviously, as you said, we had a round table earlier in the year, but we've done a bit of thinking since then, haven't we? So can you just tell us a little bit about what we've done? 

Chaitanya: Yeah, so I think the round table generated some really, really exciting insights on where the field is, what sort of exciting research is happening in this space, and it was clear that there isn't much close to clinical implementation at that point, and we had lots of open questions around what would implementation even look like and what sort of policy and other levers need to sort of work together to make that happen. And we've sort of continued that conversation post round table with several other experts as well who couldn't make it to the round table. And, we got some really useful insights on some sort of practical matters of how to make this happen, like who needs to be involved in this, who needs to sort of think about this integration, as well as keeping in mind that the healthcare resource is limited and there are like data related capacity issues as well.

Susan: You mentioned that multiomics is only really just reaching clinical practice in some places. But, I mean, what kind of outcomes could we see if we could use multiomics, particularly from a clinical perspective? 

Chaitanya: Yes. So the exciting thing about multiomics is that it provides a comprehensive, holistic view of the biological processes happening in your body.

So someone, when I was having a recent conversation with someone, they said that genomics is like the blueprint, but what actually transpires you need transcriptomics and proteomics data to sort of really tangibly see what's actually happening in the body, not just what genomics dictates.

And so you sort of are adding additional layers of information, which would ideally help you really pinpoint what's happening in a disease and sort of really understand the underlying mechanisms of a disease. This could mean we have better risk protection for diseases. This could look like we now have better ways of informing what sort of treatment needs to be prescribed to patients and really just individualize patient care. 

Susan: Wow, that really could have quite far reaching impacts then. And you talked about diagnosis, you talked about risk prediction. Do we have any views on where actually we think multiomics could be deployed particularly effectively?

Chaitanya: Yeah. We did talk about this with some of the experts and, had some interesting insights on that. Just because AI is involved and you're sort of adding multiple layers of information, there was a question about whether we could use AI-driven multiomics to really help disease areas, whether it's in enough sample, whether there's in enough information.

So this being like rare diseases. Where, you know, like variants of uncertain significance, or VUSs, are quite common that you sort of derive from genomic analysis. And then it's the question of whether we can use transcriptomic and proteomic to supplement it and really pinpoint what some of those functionalities of those variants could be.

Another area was like women's health, just because we know so little, perhaps multiomics is at one area where we could do, sort of, a data led analysis to try and figure out some of those women specific aspects as well. 

Susan: Great. That sounds really positive. You've touched a bit on the point about data, both obviously multiomics creating lots of data and AI maybe being the way that we can help to start managing it, harnessing it, making the most of it.

But I guess, what are the first steps we might need to overcome? Because actually you need good data sets. They need to be high quality. So what is, what would that, what would we need in a kind of multiomics perspective to make sure we have the data sets that are actually gonna be fit for this purpose?

Chaitanya: Yeah, so that's an excellent question, Susan, and one, which is very difficult to answer. 

Susan: Sorry about that. 

Chaitanya: Because, I think at least on the research side, people are, like consortiums, especially, they're trying to create databases, be it for proteomics, transcriptomics, and genomics, and really amping up those resources.

And I think that's like an excellent starting point. Especially if you're doing, like, a data led analysis. You sort of need these large sample sizes to make inferences. But when that comes to clinics, it's a bit tricky because if you already know what information you're looking for and if you have all these databases, you can refer back to it. But to do like a data led analysis at the clinical stage, that might be, uh, sort of not possible. 

Susan: And is that not possible, because the data's not in a way that it's consistent. It has the same kind of key… You can try and transfer across. 'cause obviously what we're trying to do is piece together, as you said, genomics, transcriptomics, other omics which might all come in a different format. So there's a kind of. Simply formatting and kind of minimum data challenges. 

Chaitanya: Yeah, I think, yeah, it's just like the way things are set up right now. It's like you have bioinformaticians, excellent bioinformaticians working in the NHS and clinical scientists. Who can do that for you. But then it's like, are we there yet to sort of make conclusive?Are there tools to support the these analysis? Are there resources and pipelines set up to do these analysis? Do we have the computational resources to support these analysis? That sort of things become more prominent as well when you sort of look into a clinical context and it's, and sort of understanding that when it's not like a target analysis, things will take time.

And how much of that interest to us, like a more discovery led science as well. And it's trying to figure out. How much of that is clinically valuable versus something that you will refer to later? Just collect data and refer to later? 

Susan: Okay. I think that's, it's really interesting, isn't it? And I think as you said, it speaks to complexity and it almost speaks to, I think the more we talk about the world of multiomics and what it could be, it has so much potential, but because it could be many things that does create the tension that we don't have a common view of it, and therefore we create multiple things, which are all variants of omics. Yeah. But therefore then have slightly different configurations or data sets. And, and while that could be great and really flexible, it creates complexity in what's already quite a complex ecosystem. Would that be fair to say? 

Chaitanya: It does. Yeah. I think so. Yeah, and there's, there's like the upside that when you sort of have that level of flexibility, you can sort of pick and choose what you need, sort of, sort of tailored to your needs.

For example, in a research context. You need all that data and you want to put them together and you want to get the results out of it. But perhaps like for a clinician, they do like a genomic test. It doesn't give you results that they want, that they want to add the transcriptomic and proteomic layer.

So it's not quite like simultaneous analysis of everything. It's like layered on. So there's like multiple ways of doing this, and you are right, it does create that sort of additional complexity because of it. 

Susan: You've talked about the fact that multiomics is sort of just getting to clinical practice. Where is it most well developed? Where could we sort of start to see it actually having an impact, do you think? 

Chaitanya: Yeah. So I think this is something I touched upon in our previous briefing, which was out of the round table and there were some interesting projects, which are sort of raised there. So one was the Southampton Inflammatory Bowel Disease Project.

So it's where you sort of collect real time data from patients via a swallowed camera. They demonstrated how AI-driven image processing combined with genomic data could become a standard part of everyday clinical practice, which I thought was just marvelous. And then there's other companies like Nightingale Health and they have a metabolomics platform where they're exploring clinical applications, particularly, in predicting health risks and monitoring disease progression.

Then we found a few other cases, where they're using it for infectious diseases, which I think could be quite big as well, just because it's such a big issue these days. And the most exciting, I thought, was the sort of mRNA genomic cancer vaccines where they were using both genomics and transcriptomics to inform, to personalize vaccine design essentially.

And there's like MHRA involvement already in trying to develop regulations around this. So that was quite promising. 

Susan: Thanks. I think it's really useful to understand what that could look like and to, I think it reiterates the point that multiomics is not one thing. 

Something you've referred to is the role of genomics within multiomics. Obviously, PHG Foundation does come from a kind of, you know, a heritage of genomics. 

Chaitanya: Mm-hmm. 

Susan: But I think we recognize it doesn't have to, genomics isn't, isn't a mandatory component of multiomics. Is that, would that be fair to say? 

Chaitanya: It’s a bit tricky because, it’s like when you think about service delivery, genomics is well established and whereas other omics might use like specialized centers and are quite expensive at this point, which might change in the future, but right now it's very sort of genomics, is much more popular and easily sort of serviceable. In terms- 

Susan: So does that mean therefore, the most obvious way to perhaps expound multiomics is to use that genomics infrastructure as a base. I think, is that what you're saying? Which makes sense from a kind of practical perspective. I guess my question would be, is that always gonna be the best answer? 

Chaitanya: Yeah, I think there isn't one clear answer to this, just because it depends on what the test is and what the disease is as well.

For example, if there's like a separate multiomic test and that comes out as like a POCT of some sort. 

Susan: So that's ‘point of care testing’.

Chaitanya: Point of care testing, or some sort of, you know, more assay based, proteomic test. You don't necessarily need the sort of genomic service delivery to go, for it to go through.

But for other cases it might be the case that you do genomics as the first line treatment and then as the need requires, you sort of add on other sort of specialized capabilities to it. So there's like several different ways it can go and it's both exciting and daunting at the same time because of that.

Susan: Thanks. I think that's really helpful. And I guess that just speaks to where, you know, from our perspective we've come from, which is, we see a role for genomics, but we don't see that as exclusive.

We bring together the different, can we find a way for different interested stakeholders, clinical groups, and others to actually work on this? Because the opportunity is there. It's growing, as we said, there is some pioneering work going on. Yeah. But there's also a risk that we end up with quite a piecemeal approach, isn't there?

Chaitanya: Yeah. 

Susan: Could you just tell us a little bit more about what might be some of the considerations, so, the key stakeholder groups might have around multiomics and AI?

Chaitanya: Yes. So we've sort of broadly recognized a few stakeholders involved in this space and what AI-driven multiomics could mean for them. Um, so starting off, we have clinicians where I think we've recognized that having some sort of clear guidance is quite essential to help them make informed decisions from any sort of multiomic tests they use.

Susan: Yep.

Chaitanya: 'cause existing strategies might just not apply or might not sort of sufficiently cover things arising from these new tests. And then there's also patients who because of the complicated results that might arise out of it. There might be a need for additional support to help understand some of these results in the genomics context.

People might rely on genetic counselors, for example, to sort of really understand some of these VUSs and uncertainty in any sort of diagnosis that they get. And we wondered perhaps something equivalent like a multiomic counselor might come in handy at some point just to make sure that patients are comfortable in understanding what the test is, what it's for, what it means and what they should sort of do about it really.

And then obviously there's like industry and tech developers and for them, as with any other technology, there's perhaps more clarity needed on the procurement and deployment cycles within the NHS and just trying to understand what a good product is to sort of deploy into the NHS and what type of evidence is really required to show that this, that any sort of thing, technology that develop is has clinical validity.

And then for regulators as well because there's AI and now there's like multis, there needs to be like some sort of evolving framework to really sort of try and capture the sort of tests are sort of deployed in this space and we completely understand that because we don't have a clear test, regulation at this point would be very tricky. But it's something to sort of keep in mind is like, how would you sort of construct like a framework that might perhaps need to sort of evolve with the technology. 

Susan: And as we said, obviously, I mean, you can see how it is gonna be challenging from a regulatory perspective.

Chaitanya: Yeah. 

Susan: Different tests used in slightly different ways. Overlaying with AI. I mean, I think I’d be scratching my head right as a regulator. 

Chaitanya: Yeah, yeah, yeah. 

Susan: Okay, great. We've obviously talked about multiomics quite generally about some of the opportunities, the sort of the key stakeholders, but where does it sit within the context of what's perhaps happening in the health policy world?

We recently had the 10-year plan for health in England. Is that an opportunity to support multiomics? 

Chaitanya: There is definitely avenues there which could be used to support AI-driven multiomics. One, just being like there is very keen interest in adopting AI into the NHS  as well as there's like quite a few provisions being discussed to make genomics, uh, and sort of improve the genomic service.

And I think perhaps now would be a good time to sort of, not just limit it to genomics, but also sort of make space for other omics as and when they arise.

Susan: I guess also within the 10-year plan, there is quite a big commitment to the kind of the moving data into a digital format. 

Chaitanya: Mm-hmm. 

Susan: Which again, should help. That's obviously a long-term ambition, so I think with many of these things, isn't it? 

Chaitanya: Yeah. 

Susan: The ambition is there, which is great. Understanding the reality and then how that then interfaces with something like multiomics as it evolves. It's gonna be an ongoing conversation, like many innovations.

Chaitanya: Yeah.

Susan: You've talked a lot about the kind of the clinical potential, but often with innovations while glorious, they can be a bit ahead of their time. Is that a risk for multiomics? 

Chaitanya: Well like we sort of discussed here there isn't quite yet a product ready for market, but I think the word of caution, as is with other areas, is to make sure that any sort of solutions you develop, it's not technology led, but it sort of really, really addresses a clinical need.

So you sort of really need to keep the patient at the forefront of your sort of technology development and that could mean involving some of these clinicians and patients whilst you could develop these solutions because they would be able to offer insights into where the unmet areas are and what sort of solutions are actually needed, which would really inform sort of user design process of these technology as well.

Susan: We've talked about areas of unmet need. You know, patients who really want a diagnosis or an understanding of what's going on with them, or broader themes like women's health, which are often poorly represented in approaches. Multiomics obviously has that opportunity. But I guess my question would be, often those unmet need areas don't have greater dataset in the first place.

So can we hope that multiomics will work if the data isn't there to support it? 

Chaitanya: Yeah. I think for such cases… So, I would say that multiomics is offering a sort of granularity. But if the issue is sample size, there's only so much multiomics can do. But I think what people were speaking to is the fact that AI is perhaps one solution that could really leverage the minimal amount of data sets to sort of identify some of those patterns.

And there's this whole suite of synthetic data that's sort of coming up. Especially for rare diseases. It's quite interesting to see how that would help address some of these like low sample size issues and perhaps multiomics is one of those data resources that might be used to generate these type of data.

Susan: Okay. That sounds really positive. And I guess, actually, sorry, thinking about it, the more we talk about it, if you can use multiomics to show the potential, then you can make the case almost to then increase sample sizes to improve data collection. So it might at least show us the light, show us the direction we head, and then we can think about how to actually address the unmet need through better data. You know, by knowing that if we get the data, we can probably help. It strengthens the case for getting that data. 

Chaitanya: Yeah, definitely. Yeah. Yeah. 

Susan: We've touched on various elements around multiomics. If there was a point of optimism for the field right now what would you say you are feeling most optimistic about?

Chaitanya: It's just the potential of what it can achieve, which is a sad thing to say in a way because it's not concrete enough. But there is like immense potential in what it can achieve. Looking at diseases in new ways and discovering new sort of treatments for it can really sort of help navigate some of those gaps where people are just stuck in diagnostic or disease.

I think that's like a message of hope. I would leave with for sure. 

Susan: I think that's really powerful, isn't it, to think about either areas of huge unmet need or places where, as you said, which were a bit stuck. 

Chaitanya: Yeah. 

Susan: And this, this idea of offering a fresh perspective, a new way of perhaps interrogating information that we hadn't thought about.

Chaitanya: Yeah. 

Susan: Is in itself you're, you're right, I think really exciting. I think just to counter that, you know, is there a risk or a watch out? You'd just say, we need to sort of flag. 

Chaitanya: I think the main risk I would say is sort of losing momentum that's sort of been built in this space. There's so much research happening, there's lots of enthusiasm.

Everyone we sort of spoke to was quite enthusiastic about it, and it's just making sure that when the time comes that something is quite ready for clinical adoption, that there is clinical infrastructure and sort of strategies in place to really adopt them easily. And it's not that we wait until that point to then start thinking about what the infrastructure needs should be, what sort of guidance should be provided to clinicians and patients.

And it's sort of, making sure that that sort of, we can make this process as seamless as possible. 

Susan: Yeah. I think that point about inaction's a really good one, isn't it? 

Chaitanya: Mm-hmm. 

Susan: And it's so easy to not do something. 

Chaitanya: Yeah. 

Susan: But actually the risk of not doing something can often end up with greater harm or worse outcomes.

Chaitanya: Yeah. 

Susan: And I think it speaks well to your point around how we need to bring together those clinical stakeholders, because collectively, they will actually give us the granularity and the richness to start to think about this. I think what comes through with all of this work is the role that it's not one individual group or one individual organization that can make sense of this. It is a multi sort of stakeholder area, isn't it? 

Chaitanya: Yeah, and I am careful in sort of bringing up clinicians here because they are already so burdened with all the work they do under high stress situation. And it's not necessarily that all clinicians will be adopting these solutions either. There'll probably like a few sort of closely tied in with academic research who might be the first to sort of bring some of these results.

It's just like being aware of that as well. 

Susan: Yeah. I think you make a really good point. We are not trying to implement this across every clinician and every part of a health system. It will be targeted, as you said. I think the… Often these things start closest to the kind of the research active centers, just because that interface is easier.

But I think even more broadly, we'd recognize that it will have a specific role as a kind of part of perhaps a more personalized health offer. 

Chaitanya: Yeah. 

Susan: Deployed as appropriate. 

Chaitanya: Yes, definitely. Yeah. 

Susan: Chaitanya, thanks. I think it's been a really interesting chat around the intricacies and interests of multiomics and AI.

I don't think it's the end of this yet, is it? I think we see this as a conversation starter. 

Chaitanya: Mm-hmm. 

Susan: And, you know, a scope to explore work in the future, and also to think about how it interfaces, as we said, with sort of government policy and other changes ahead. 

Chaitanya: Absolutely. We did produce this with the intention to be a conversation starter and we definitely want discussions to continue in this space and we'll definitely be keeping abreast of everything that's happening and there's so much happening. So it's just like a very exciting time at the moment. 

Susan: Excellent. I look forward to a future chance to discuss this with you.

Chaitanya: Me too. Thank you.

Ofori: That brings us to the end of the episode. If you would like to find out more about what was discussed, please email us at intelligence@phgfoundation.org. Thank you for listening, and we look forward to bringing you a new topic in the next episode.