#260 Ash Anwar: How Molecular You Uses AI for Early Disease Detection

0:00The datasets we started with are exactly what you said, labeled datasets that usually accompany a lot of the published information too. So definitely the deep scientific literature, clinical literature, we're able to acquire the data sets of patients either from the authors or the publications that have used this. And a lot of biobanks will also have this information available as well. So being able to use that data as a basis to kind of develop the models and then also battle test this and see how accurate the models are at predicting disease as well with validating through real patients. The research is to curate and identify novel biomarkers that we want to incorporate into our platform.

0:43We are also constantly evolving the different disease areas and biological mechanisms that we're including into our platform as well. Build the future of multi-agent software with Agency, A-G-N-T-C-Y. The Agency is an open source collective building the internet of agents. It's a collaboration layer where AI agents can discover, connect, and work across frameworks. For developers, this means standardized agent discovery tools, seamless protocols for interagent communication, and modular components to compose and scale multi-agent workflows. Join Crew AI, Langchain, Llama Index, Browserbase, Cisco, and dozens more.

1:34The agency is dropping code, specs, and services. No strings attached. Build with other engineers who care about high-quality multi-agent software. Visit agency.org and add your support. That's A-G-N-T-C-Y dot O-R-G. Join other engineers who care about building the internet of agents. So go ahead. Yeah. Sure. So I'm Ash Anwar. I'm the co-founder and senior director of scientific research and development at Molecular U. My educational background is rooted in science and business. So I did my undergraduate degree at Simon Fraser University in molecular biology and biochemistry. and then I went to the University of British Columbia to do my graduate degree in experimental medicine where I did research in neurological conditions like Alzheimer's disease, multiple sclerosis and then I went on to Harvard Med School to do my research fellowship in ophthalmology.

2:40I was really interested in transitioning from neurology to vision disorder so my research project at Harvard was around age-related macular degeneration which is the leading cause of blindness and older adults. And once I was complete there in Boston, I decided to come back to my hometown, which was Vancouver, and look at opportunities to transition from academia into more of an industry scientific role. That's when I met Rob, and he mentioned that we are starting up this company in experimental, in looking at translating scientific information and getting into the hands of individuals. And that's what my entire program in experimental medicine was all about, to go from bench to bedside, as they call it, go from scientific research to the bedside.

3:35And so I got really passionate about that and we joined the company, got it off the ground and everything. And so I joined Molecular U and that was back in 2000. 15, I believe. And yeah, and you know, we're bringing all that information together, I was really able to work on this information from the ground up and really help build that. And while I was at Molecular U, as I was growing in the company, growing the team, and we were growing the organization, I saw that there was a need for translating not just the science information, but also understanding how science can play a business role in industry as well.

4:17So I also ended up doing my MBA from the University of British Columbia as well to help wrap up on that aspect and get that knowledge base of business too. Yeah. And the premise is that there are biomarkers out there And now we have enough data and enough powerful enough algorithms and compute to see what effect those biomarkers or what those biomarkers can tell us about the emergence of disease and longevity. Is that right? And I got to say to listeners, I interviewed Eric Topol, who's a well-known doctor in the AI space, about a book he has coming out. I think I mentioned it to you last time we spoke, called Super Agers.

5:22And it's specifically about this, about that there exists this technology that can be applied to the data regarding these biomarkers and that we should be able to predict, you know, the onset of the major disease classes that limit longevity. and if you can predict them, then you can address them, even though we know how to address them anyway, but people don't necessarily follow doctor's advice without a deadline, so to speak, in a very real sense. So can you talk about, so I'm going to run this episode in, you know, back to back with the one about Eric, because you guys are doing precisely that.

6:20Can you talk about

6:26how molecular U works, first of all? Yeah, absolutely. I think the premise is exactly what you mentioned. We are looking at the early indicators of disease. Traditional medicine with the biomarkers that they're looking at really is in this diagnostic classification. So what the physicians are looking at are the biomarkers that signal that disease and damage has already occurred. Symptomatically, the patient is exhibiting certain symptoms of the disease, basically. And so that's really, if you think about a flow chart of how disease works, that's really at the end. When things have created damage, something is wrong in the system, then the patient goes to the physician and says, hey, I think I feel bad.

7:17They look at the biomarker and say, OK, you have this disease, potentially. I'm going to give you a medication to correct the factor that's happening or the symptom that's usually treatment of the symptoms that's occurring. What Molecular U is looking at is we're looking at the early stages of that disease. And this is where our bodies are actually quite resilient. So there might be damage occurring in our bodies, but our bodies are essentially maintaining what's known as homeostasis, maintaining a state of normality as best as it can while the damage is still occurring. and over time what happens is this damage can accumulate and if you think about like dominoes like the first domino may start to fall but it's only until you get to that last few dominoes where the entire thing collapses and then you are in the full throes of the disease and so we're looking at those early root causes of the disease this is where these biomarkers come into play and that's where not only are we getting an indication of the specific early indicators of the disease, but a lot of times diseases work through multiple different pathways.

8:26So developing diabetes or cardiovascular disease, for example, one individual may go down a certain biological pathway to develop that disease, but another individual might go through another set of biomarkers that leads down to the same diseases. And this is where molecularity looks at the combination of those biomarkers and is able to inform that, okay, this is the path that potentially can lead to a risk of having that disease. And as you mentioned with the super agents, if we could look at those biomarkers that impact our longevity, impact our ability to maintain a disease-free life, we can actually target those during those early cases of damage and then reverse at that stage where things are very reversible and get us back on track to a healthy longevity state rather than waiting until symptoms occur and that's when you've had so much accumulation of damage it becomes much much harder to get back to a healthy state right so uh and the idea is that you you track a patient over time this is a longitudinal study where you would have a blood test annually and see how the biomarkers are changing.

9:41Is that right? Correct. Correct. Yes. I mean, that's, and that's really powerful because looking at how we uniquely change over time and how our biomarkers are dynamically changing with our environment, the stresses we have in our environment, what we eat, what our exercise habits are, things like that that is all going to have a significant impact on our biomarkers and being able to longitudinally track how we change and how the the dynamicism of that works gives us an unparalleled resolution into how our bodies are are trajectoring over the the the health landscape essentially right and presumably uh this is you know you train the model uh and then you back test it with known data to improve the model or optimize the model, fine-tune the model, is how effective has this process been in identifying when you look back over time?

10:45It's been very, very quite effective, actually. I mean, being able to retrospectively look at the data when we understand the endpoints of where an individual has gone towards, we're able to essentially trace back the path that certain biomarkers have taken to go out of that normal range. And we've been able to look at the different biological processes that have gone awry in the early stages. So looking at having a model that enables us to predict a certain chronic disease, for instance, and then looking at data where the individual or the group of individuals has developed the disease that enables us to like be able to identify those different tracks not just one but multiple different tracks and at what stages these biomarkers have have gone out of the normal ranges and being able to feed that beta data back into the model has increased the accuracy in our predictive value of these models so it's been a really balloon for for that Yeah.

11:47And how many markers do you track? As of now, more than 250 biomarkers. And these biomarkers are the metabolites and proteins. So metabolites are essentially the chemical byproducts of the chemical reactions that occur in our bodies. So anytime we exercise or eat or have stress, if we're not sleeping very well, the chemical reactions in our body will result in these metabolites that are very dynamic and will change. And then also this other omega as well, which we'll analyze, which is the proteins. And these are the structural and functional elements of our body. So these are the things that make up our muscular skeletal system.

12:30proteins are essentially a big part of the enzymatic components that enable these chemical reactions to occur as well. So it's 250 biomarkers plus right now, and we're looking to increase to around 800 plus within this year as well, which will give us a wider panel. And you said that that you only need a vial or two vials of blood to track? Is that possible? 800 biomarkers? Yes, yes. So it is a full vial of blood, just like you go to the doctor and they take one vial of blood. The technology we use is mass spec, so it enables us to be able to essentially use this sample and go through a high-throughput panel that enables us to analyze a lot of these biomarkers at the same time.

13:27So this is the process that gives us the ability to increase the number of biomarkers while still not exponentially increasing the amount of collection that we would need to do for the patient. And how frequently do you need to sample the blood to track changes in the biomarkers? Yeah, this is a really great question. I think it's a great question because the answer is it depends. It depends on the situation. Very often, I think a very common way of doing this longitudinal assessment really is an annual checkup where the individual comes in and does essentially an annual test to look at where they are in terms of their health state.

14:13And then in the next year, they will go and do another test to see has there been changes? What is the delta of the changes of these risk trajectories that we're looking at in terms of these different chronic diseases that we're all put on? So that's, I would say, a very good default situation. However, there are a few, I would say, a few cases where if an individual is already been diagnosed with a certain disease, maybe they have diabetes already or they've had a cardiovascular event in the past, like a heart attack, that's when having a shorter or more increased frequency of testing will lead to the benefit of being able to track even more quicker changes that may occur and actually taking, doing interventions even quickly to have even maximum benefit in those specific individuals.

15:05Yeah. You know, you showed a presentation when we first spoke and it had a case of a woman who you had followed for some time, I don't remember how many years and in the visualization that you showed you could see something was starting to happen around 2019 i think by 2022 there was clearly uh something wrong and she subsequently or or as a result had a full body i think it was a ct scan or an mri i'm not sure which yeah it was a ct scan Yeah, and they identified a lesion on her pancreas. It turned out to be malignant, but it was stage one, and so they were able to operate. I don't know if she had chemo or radiation, but that I'm sure resonates with a lot of listeners.

16:06I just had a friend die very suddenly and unexpectedly of pancreatic cancer, And one of the problems is that it's asymptomatic until it's stage four and then it's too late to do anything. So, yeah, can you talk about that? Was that unusual? How often are you able? Well, first of all, how many people are you tracking? We're tracking. We're talking about in the thousands of individuals right now and exponentially increasing. now that we've done our launch into the US market as well. So we have a lot of clinics on board and are deploying this test to their individual patients as well. So it's increasing a lot, the number of individuals we are tracking at the moment.

17:01And yeah, I'm very sorry to hear about your friend. I mean, cancer is a very devastating illness and very often, especially in the case of pancreatic cancer, It is very hidden until the very end where the cancer becomes really aggressive and really there isn't a lot of interventions that can occur. We were actually quite lucky with this patient as well. I mean, she's been with us for a very long time and has been doing these annual testing as well, more so from a curiosity standpoint and wanting to be in the space of longevity. She was she's 60 years old. So she really wanted to think about like, oh, how do I have a good quality of life well into my later years as well?

17:48And I want to make sure that I optimize my health. I want to make sure that I understand what is the lifestyle that I need to maintain to keep this high quality of life or even optimize this, the quality of life that she has. And it was during the time of COVID, actually, where she started complaining about fatigue, excessive fatigue, not being able to work as effectively as she did, shortness of breath, things like that. So she approached her doctor around some of these symptoms, and they were very nonspecific, and they weren't really able to indicate what could be happening other than maybe overwork or something.

18:31However, in her molecular U test, this is where we have the ability to look into it. And we saw that there were quite a number of biomarkers that I know from my biochemistry background that these are indicators of cancer environment, like a tumor type of environment. And several of these biomarkers were orders of magnitude higher than what we would expect in the normal range for a healthy individual. And this is what really flagged for us that, hey, something is really going wrong here. And taking a deeper dive into it, we noticed that these biomarkers are related to cancer. And there were a number of, you know, the thing with biomarkers as well is they follow a certain path of disease mechanisms that may overlap with different diseases.

19:19So we noticed that with pancreatic cancer being one of the risk factors, there was also ovarian and breast that may also be a risk factor as well. So we relayed this information to her doctor to say that, hey, we're seeing cancer biomarkers that are relevant in her profile. And kudos to her doctor. He was very receptive to this information. he spoke to her there's a family history that that she had with a sibling who passed away from ovarian cancer and so the doctor decided that hey we're going to do a ct scan to see what's going on and that's when they found the the lesion uh did a biopsy and and that's when after doing the biopsy of the lesion they noticed that the biomarker for pancreatic cancer was a thousand times higher and they were like okay this needs to be surgically removed luckily it was at a stage stage one where the cancer hadn't became aggressive yet and so they were able to remove it just through surgery no chemo was needed to be done and it was just this after the surgery recovery that's all that she needed to do now she's doing fine and we've done follow-up tests to also verify that all these cancer-associated biomarkers are now back within normal ranges.

20:38And now it's just monitoring. Now it's making sure that once an individual has had cancer, there's an increased risk of cancer coming again. So we just want to make sure we monitor her. And any time we see another signal or something, we know right away, we can put her on that clinical path to getting that normalized. Yeah. Build the future of multi-agent software with agency, A-G-N-T-C-Y. The agency is an open source collective building the internet of agents. It's a collaboration layer where AI agents can discover, connect, and work across frameworks. For developers, this means standardized agent discovery tools, seamless protocols for interagent communication, and modular components to compose and scale multi-agent workflows.

21:32Join Crew AI, Langchain, Llama Index, Browserbase, Cisco, and dozens more. The agency is dropping code, specs, and services, no strings attached. Build with other engineers who care about high quality multi-agent software, visit agency.org and add your support. That's A-G-N-T-C-Y dot O-R-G. Join other engineers who care about building the internet of agents. You know, there are a number of companies that are offering, you know, screening, biomarker screening now. and it's difficult to know which really have the models capable of doing the analysis which biomarkers are meaningful uh eric topol was uh i mean i didn't had not spoken to you yet when i spoke to him but he was cautious about a lot of these companies because he wasn't confident that they're following the correct biomarkers how frankly how do I mean have you published a lot of this I mean how do people how do you engender confidence particularly after you know blood biomarker scandals in the past that that this is real?

23:15Yeah, yeah. No, of course, I think that's a very, very valid question. I mean, even here, us at Molecular U, we're only scratching the surface when it comes to biomarker testing, too. I mean, there's thousands of biomarkers, if not hundreds of thousands of biomarkers in our body, and they're relevant to a lot of different degrees, and it's really about picking the biomarkers that we believe have the highest value. And even then, it's going to have value only in specific, perhaps niche areas of interest as well. We have published on a number of our studies that are looking at these biomarkers and models and predictions of disease.

23:59And a lot of this is also classified into different, I would say, different omics. a lot of the early companies had focused on genetic testing because that was a really hot topic. You know, could we look at someone's genetics and be able to predict diseases that they are predisposed to down the road? And so a lot of this personalized medicine and precision medicine really bore out of that genetic testing. Unfortunately, you know, one of the things that we've seen with genetic testing is that there's a lot of misinformation with the genomic information as well. And secondly, the genes themselves don't exactly, like being able to predict disease from genomics alone is not a very strong way of going about this process.

24:50Someone could have a genetic predisposition, but that could potentially not be expressed and the individual may not end up having that disease. And so where molecular you is really uniquely positioned is we actually don't do genetic testing. We are looking at the dynamic biomarkers in the blood that change as things progress just naturally through healthy aging, but also through any of the stressors or any life events that an individual is having as well. And these are the biomarkers that would directly affect the mechanisms that govern health and disease. The other aspect of molecular that is, I would say, uniquely positioned is that the biomarkers we're looking at, first of all, is a very, I think we have the largest panel of biomarkers that we are looking at, but we also have an active scientific research team where we're constantly surveying the literature to identify any new biomarkers that are of significant an interest and high value, and then working towards experimentally bringing that into our panel to incorporate and boost up our ability to make predictions on a number of different chronic diseases as well.

26:04But again, I'm not going to say that we are catching everything, because there's a lot of areas that I'm sure we have gaps. We're working towards identifying and being able to fill those gaps as well. Yeah. When was Molecular U, I'm sorry if you've already said, when was it founded? I believe it was founded in 2014, 2015. I see. And so you have an early cohort that you've been tracking since then. Yeah. Right. Yes, yes. And the analysis, is there enough backdated data on various biomarkers tied to patients that you can make some assumption about how they're tied to various diseases? Yes, yes. And we when we've been working, we call them the pathfinders because they were the initial cohort that were that we used to help us validate a lot of our models or systems and being able to measure the accuracies of what we're predicting.

27:28and absolutely like being able to track them over such a long time and to see kind of the endpoints that have occurred has enabled us to retroactively go back and look at the data and understand the trajectories of the different common editorial biomarkers that have resulted in different health outcomes for these individuals. So yeah, it's been really great to be able to have access to that information and being able to almost, I would say, simulate what another patient with a similar profile biomarkers could be potentially along that track of disease, essentially. Yeah. What is the model? What are the models that you're using?

28:11Yeah, it's a great question. We have a number of different models that we use depending on the disease and the type of biomarkers that are associated with it. Essentially, it is like supervised machine learning, right? So we supervise machine learning classification model. We have biomarkers that enable us as a signature or a set of features to be able to classify whether a person is going to be in the class of healthy class group, or are they aligning with the important features that we've identified in the biomarkers that put them in the class of disease, essentially. um what the specific type of model that we use again it can change uh support vector machines is something we use quite quite often we have a publication around that where we've used uh support vector machines to develop a model for cardiovascular disease to identify patients at high risk of coronary artery disease so we i've actually published that with the team back in I can't remember the year, but 2019, 2018, perhaps around there.

29:21And so a lot of those models, as I mentioned, could be support vector machines. We will use neural nets as well. And as the machine learning models or AI has become more and more sophisticated, we've adopted more and more of the more cutting edge tools to see if we can enhance the prediction of the models. Can we make the model better? Can we incorporate other data inputs as well? Currently, the team is looking at incorporating not only so at the current moment, we ingest a lot of information from the scientific research literature, which is constantly increasing. Every year, I think there's more than millions of papers that are published every year.

30:03So we have an AI engine that's able to scour all the scientific and clinical literature and pull out the key pieces of information for our scientists to evaluate and consider incorporating into the model. So that's one data input that we use. Second data input that we use currently as well is this real time patient data. As we go through patients, they input their information as they go through their longitudinal health journeys, we're able to use that information as well, that data to help make our models better as well. But especially this year, the third aspect of data that we're really looking forward to is how do we integrate other data inputs?

30:42Can we look into integrating electronic health records into our data? Things that the clinics are routinely taking a note of every time they see a patient, whether that's doctor's notes, electronic health information, perhaps in the future wearables as well. But this is where things become a little bit complicated because there is a lack of data standardization when it comes to information and wearable information. So that's a big challenge for us right now to incorporate the data. But it's kind of an active area of research. And the more modalities we can put into the models, hopefully that will enhance our ability to predict this even better.

31:26Yeah, and the supervised learning takes an enormous amount of labeled data to train an accurate model. What were the data sets you started with? The data sets we started with are exactly what you said, labeled data sets that usually accompany a lot of the published information too. So definitely the deep scientific literature, clinical literature, we're able to acquire the data sets of patients either from the authors or the publications that have used this. And a lot of biobanks will also have this information available as well. So being able to use that data as a basis to kind of develop the models and then also battle test this and see how accurate the models are at predicting disease as well with validating through real patients.

32:25yeah where and where are you on the product journey are you has this is this fully productized is it still in the research stage yeah we have uh we have a product that out is out in market right now and we believe has a good product market fit especially with clinics that are and institutions that are looking at deploying this in a longevity and and wellness wellness kind of space. We also work with a lot of the product is also very useful for clinics that are in, I would say, call them specialty clinics as well. We've done work with neurology clinics, for example, where they have patients that have certain varying degrees of cognitive decline, and they want to do an analysis to understand is it a cognitive, perhaps a cognitive impairment that is coming through a nutritional issue or a metabolic issue, or is this something along the lines of a neurological component like Alzheimer's disease.

33:32So our product is developed as well for these specialty clinics that want to look at specific niche disease areas as well. So if you have a good product side that is able to work b2b with the clinics uh we do have a d2c angle as well so patients if they're interested are able to directly order the tests and get the results and and have an actual platform that they're able to look at all these biomarkers understand what these biomarkers mean for their health and and uh trajectories for different diseases as well And this is the very commercial side of Molecular U. Where I'm also really actively involved is this very deep research side of Molecular U as well.

34:18And that really feeds into the product, right? As we develop things on the research end and validate it, we can actually feed it into the product and into the commercial space. But the research side of it is really involved heavily with a lot of academics, for example, that are involved in research into different diseases, looking at novel biomarkers. And we provide our ability to our panels to look at these, but also our machine learning and AI capabilities to help them ramp up their modeling efforts on the research side as well. yeah uh as i said i've seen a lot of uh companies uh doing similar things who who do you regard as your biggest competitor it's a good question very good question uh function health i would say comes to mind um they've just closed a big round as well and and are if i'm not mistaken i think are valued at in the billions of dollars.

35:20So I think Function is probably our biggest competitor. The unique differentiator, again, with Molecular U is that Function focuses a lot on a lot of standard clinical tests that are routinely available to doctors already. However, they're able to take those standard tests and combine them and come up with these health areas of interest that the clinicians find useful as well. Where molecular use is very uniquely positioned is we go much more beyond just the standard clinical tests and look at the plethora of these biomarkers that are more pre-diagnostic and really giving you that early indicator of disease before things are going awry.

36:06But yeah, function, I believe, is our... If I was to pin my hat on our competitor, it would be function health. Yeah. And then where are you in funding? I mean, you mentioned function just had a big round. How much have you guys raised? It seems like there'd be a lot of money for this kind of activity. Yeah, no, absolutely. I think we're in the midst of our Series A right now. In terms of the actual number, I don't know the actual number that we've raised. But I can say that, you know, comparatively to what Function has managed to close on, I would say that investors looking at our technologies are getting a significant discount in terms of the potential for where we're at.