Transcript of Episode 68: Caspar Barnes

Disclaimer: Transcripts are automated and may contain errors.

Grant Belgard: Welcome to the Bioinformatics CRO Podcast. I’m your host, Grant Belgard. Today, we’re speaking with Caspar Barnes, founder and CEO of AminoChain, a startup marrying biobanking and blockchain to make biospecimen sourcing transparent, ethical, and efficient. We’ll explore what AminoChain is doing today, how Caspar’s path unfolded, and the advice he has for the next wave of biotech builders. Caspar, welcome to the show.

Caspar Barnes: Thank you so much for having me, Grant. Excited to be here.

Grant Belgard: How do you describe AminoChain to someone who you meet in an elevator?

Caspar Barnes: Yeah, absolutely. So AminoChain is a decentralized biobanking protocol. It’s an infrastructure company that connects hospitals, biobanks, pharma companies, and other users and actors in the life sciences industry. And it allows any number of decentralized healthcare applications to be built on top. The first app that we are building on this decentralized biobanking protocol is a biosample marketplace that we call the Specimen Center. And how it works is we want to turn donated specimens into non-fungible tokens. And then we let those digital assets get listed onto a marketplace. And we let life sciences companies license these biospecimens for research use. And we can encode rights into the NFTs that represent broad consent from the patients and royalty rights back to the individual donors, and perhaps even MTA and licensing conditions of the biosamples.

Grant Belgard: Which steps in today’s biospecimen procurement pipeline are most painful?

Caspar Barnes: Oh, there’s so many. Where do we begin, right? So like today, in the United States alone, there’s around 2,500 biobanks out there. And a biobank, for those that don’t know, is a really big fridge filled with donated cancer samples mostly, but all sorts of other disease tissue used for research. And across these 2,500 biobanks, there’s roughly 200 million retrospective specimens stored readily available for research use. And only around 10% of all of those samples ever see the light of day. And that is because of many different reasons. We’ve spoken with hundreds of biobanks when we started AminoChain. And the most common themes that crop up are firstly, poor financial planning. Biobanks say that they’re scientists, they’re not business people. So from their perspective, they’ll go and raise grant funding, they’ll build the infrastructure to store specimens.

Caspar Barnes: The second they start collecting samples, they’ll just go back to doing research. And they don’t think about access policies or governance on the samples or distribution policies or cost recovery models. And so effectively, poor financial planning is the main reason why these biobanks are unsustainable resource. The second thing is searching for samples is really difficult. So today, if you’re a scientist and you want to get access to a specific biospecimen from a bank, you’ll have to go individually to each bank and ask if they have what you need. They have their own bespoke access procedures where you go and try to contact the PI at the institution. They’ll go and look for the sample. And if they find it, then you’re in luck. And if they don’t, yeah, move to the next one. But there’s no real way to harmonize search across these disparate databases.

Caspar Barnes: And then the last thing, of course, is licensing the thing. Once you find what you’re looking for, you could spend on average like three months back and forth debating the conditions of a licensing agreement called material transfer agreement. And then once you’ve actually reached those types of conditions, you can sign a document and get the samples. The whole process of finding, acquiring, licensing, distributing these pre-clinical research assets from biobanks scientists is just riddled with problems. What does a three-month delay in specimen access cost to mid-sized pharma program? It can totally vary, right? But speed is everything within pharma. So your mid-sized biotech or pharma company could, especially if they raise money to build out their own library of omics data, which maybe they’re training AI on, or they’re doing target ID and validation and so on. Speed is everything.

Caspar Barnes: So three months could mean being the first person to find that insight or validate that target or not being that. And if you’re not that, then maybe that’s the entire USP of your company kind of down the drain. And so we found in many instances, people, A, want access to data straight away. And then B, if they can’t have the data, they want their retrospective specimens so that they can turn those specimens into data. And then C, if there are no specimens available, then they want actual human beings so that they can donate samples, so that they can get the specimens, so that they can turn that into data. And so it’s quite difficult to quantify, but it can quite literally be the matter of lack or death for some of these companies. And so speed is everything in the industry.

Grant Belgard: Why did you choose a permissioned blockchain rather than public chain?

Caspar Barnes: Well, we have a lot of different things to consider within the chains that we’re working with. We are actually settling on a public chain. So we most recently decided to work with a private app chain company called Syndicate, which means that we’re able to customize a little bit of the data that isn’t, isn’t visible. It could be permissioned in that aspect, but we still do settle on Arbitrum, which is an open public blockchain. And so we do leverage the security and open transparency of public chains, but we also customize to some extent the transaction data or the publicly visible metadata within a sort of private permissioned app chain infrastructure. And we straddle these two different strategies specifically so that we can be a fully decentralized protocol by settling on Arbitrum eventually. But we also tailor the app chain specific needs towards our users.

Caspar Barnes: We found before, for example, we tried to have a totally open public app on Polygon and that data being totally visible on a public blockchain made a bunch of our users nervous and the language needed to cover all of the functionalities of the blockchain and our technology and our stack in a provider agreement that we would then present towards a hospital. So a biobank was incredibly confusing to these biobankers and researchers that have never even heard of crypto before. And so for all these reasons, we ended up, you know, deciding to focus on developing our own app chain, which basically means we can customize a lot of the information that isn’t as invisible, but we still are leveraging all the benefits of being on a fully decentralized protocol, like by eventually settling on those chains as well.

Grant Belgard: Can you walk us through a typical search match compliance workflow with AminoChain?

Caspar Barnes: Yep, of course. So today, a scientist will log on to the specimen center. They can see all the different biobanks that have created the profile and they’ve listed their specimens on the search platform. We currently have a global network of over 20 biobanks, some in the European Union, some in Eastern Europe, some in Canada, and some in Africa, and some of the United States. Folks can log on and they can see the profile pages of these banks. It’s totally open and transparent. None of the specimens are blinded. None of the suppliers are blinded. Experience is meant to recreate something like Facebook for biobanks. So you can go on and connect, browse each other’s profiles and see the high level overview of the collections of each biorepository. Then you can go down to a more granular level. You can go towards a specimen level.

Caspar Barnes: And across these 20 different biorepositories, we’ve ingested all of the metadata of the collections that these biobanks have. And we’ve mapped all this metadata into a universal. So if you go into the specimen center and you’re looking for a glioblastoma from a Caucasian male, you will also get a brain cancer sample from a white man, for example. Those are the same specimen, but they just have synonyms of each other to describe it, right? So we’ve used a series of different LLMs and AI technologies to map all of these metadata against each other. And now users can come on and search for what they need and harmonize, or they can search across all these 20 different repositories. From there, they can then select the specimens and then turn inquiry. They have extra information that they want to know about the samples or about collection or about provider.

Caspar Barnes: They can add that context into a chat and send that context with the request that they get pinged directly to the biorepository. So we don’t actually get involved in licensing at the moment. We don’t involve payments. We’re just nailing the search experience for the users and for the biobanks.

Grant Belgard: How are you handling private key management for sites that aren’t crypto native?

Caspar Barnes: Right. So the specimen center in its first iteration, she doesn’t have anything on changes yet, right? So we’re slowly starting to integrate all of the app chain enabled features right now and bringing the existing transactions onto the blockchain. How we are going to do it is work closely with third party key abstraction providers, like for example, Privy.io, the company recently acquired by Stripe. And they’re fantastic. We can work with them and they can outsource all of the key management and compliance and they can provide a fantastic that abstracts away the crypto in the backend. So they make building on chain a lot easier than it used to be.

Grant Belgard: What are the best traction metrics for Amino Chain? Samples onboarded, active buyers, cycle time reduction. What do you think best encapsulates your story?

Caspar Barnes: Yeah, fantastic. It’s a good question. So the key metrics that we’re tracking is first of all, the size of the network, right? So, I mean, how many buybacks are on the platform? How many have bought into the mission of improving their visibility and improving their cost recovery? And so that the first and foremost, the main thing that we track is how many providers do we have and how many specimens do we have? And then of course, how many unique individual donors or patients do we have? That’s the main thing that we track. The next thing of course, is how many users, how many scientists are logging on, how many people are looking for biospecimens. And then the most important KPI perhaps is how many requests are actually happening on the platform. So how many channels do people log on? Do they use the full search experience? They find what they need and they send a request to the bank.

Caspar Barnes: And there’s a dual-sided approach there where you need breadth, of course, because you need to be relevant and applicable to so many different types of scientists. But often people will come and if they don’t find the specific bit of insight that they’re looking for, then they would churn and they’ll just go directly to the bank or they’ll go to another one and they’ll try to find the specimens they need elsewhere. So apart from breadth of all these different collections, you also need depth. We need highly detailed information on all of the sample donors and different collections. And at the moment, we’re only tracking perhaps, you know, 18 to 20 different fields of metadata. And some of those fields have largely unstructured data, so people can drop in clinical notes or path notes and so on.

Caspar Barnes: But the way where our search is going is into vector embeddings and into more sort of natural language processing and so on. So that means that people can come on and ask questions in natural language. And we can have, you know, agentic tools to help us find the exact specimens that they need. And we can see then if any of the insights these people are looking for lies within the data that is uploaded onto the specimen center. So all these things considered, I think the most important KPI for this would probably still be transactions or requests, because that shows that our search is providing the experience that the users want.

Grant Belgard: And how do you defend against large CROs that might try to spin up a similar platform?

Caspar Barnes: Yeah. Well, the good news is that over the last like 30 years, people have tried many times and no one has made a lasting successful platform. And the reason for this is the traditional marketplace model is to de-identify where the specimens come from and to add a markup and to force people to do payments and transactions to the platform. We are largely of the opinion that we shouldn’t be brokering retrospective biosamples. We don’t think it’s an ethical practice to add, you know, markups on top of selling diseased tissue. First thing is just the values perspective. But then the second thing as well is that if you don’t de-identify where these specimens are coming from, then there’s the risk that you have marketplace slippage. And that’s the same with any marketplace. So people would log on, they would use your platform for search.

Caspar Barnes: And then if they can see exactly where the sample is, then they’ll just go offline and buy it directly. And, you know, all CROs, all buyer sample brokers, all the big players out there, they’re forced to make money by putting the value of the transaction on the actual brokery of the tissue. So we have tried to find a way to provide value for a network without necessarily focusing on trying to extract value out of a buyer sample transaction. And I don’t think that’s actually really been done before, let alone successfully done before. So that’s our approach. If we make it totally open, and we don’t mind if you do the transaction on our platform or off platform, right? We just want to nail the search experience. Then we could end up being the platform that everybody comes back to because it is actually the thing that is more engaging for the providers.

Caspar Barnes: It does have more rare specimens on it. And there’s no reason to jump off. You actually get a better user experience finishing your transaction on the platform because there’s no, there’s no reason not to do that, right? It’s not going to be more expensive for the user. And then once we have that good retention and we have good network of both provided and procurers, we can monetize in many other ways. Firstly, with the blockchain, all the amazing things we want to do there when the specimens are protocol integrated. But secondly, even without the blockchain, just nailing the search experience is already a good, good value add for these procurers. So like, for example, when you go onto LinkedIn, you can go and scroll through everyone’s profiles in a sort of freemium way.

Caspar Barnes: But there’s these amazing, you know, added tools on top like LinkedIn Sales Navigator or LinkedIn Recruiter or whatever that people have to pay for having an extra service. But we can totally do the same thing for biobanks here, right? If you have a phenomenal search experience and you want to have automated feasibility assessments for prospective collections, you want to have a gently tools where you can drag and drop your protocol and you have an agent search the marketplace for you and so on. All of these amazing things that we want to do later, we can charge subscriptions for or other things for and we can put that onus on the actual, you know, researcher that’s looking for the specimens. We don’t have to provide any barriers towards the providers.

Caspar Barnes: And of course, the most important thing is we can take the whole emphasis on brokering tissue off of, you know, that retrospective transaction.

Grant Belgard: How do HIPAA, GDPR and other laws and regulations interact with your cross-border workflow?

Caspar Barnes: Yeah, that’s a great question. So we do have banks in the EU at the moment and what’s particularly difficult is that each country, you know, can have their own interpretation of GDPR. And so even GDPR in of itself isn’t like, you know, a standalone uniform beast that you can just address one time because each user interprets it differently. So first of all, what we do is we take in de-identified sample metadata. We take in very high level information on the collections and we make that searchable. We don’t actually get involved in the licensing and the payments of the samples as well. We highly vet all of the providers that we work with to make sure that they are GDPR compliant. It’s written into our provider agreements that they also assume the risk of being GDPR compliant and that they have the capacity to erase data if that’s what the users wish.

Caspar Barnes: And we customize the specific fields based on their interpretations of GDPR. So for example, the French banks, they think that including information on a patient that has an age above 90, for example, would be personally identifiable. So for the French banks, or under five as well, by the way. So for the French banks, we would then change the fields to say, you know, 89 plus or six under or something like that, right? So that happened, similar things happen all around the EU. And we basically meet biobank where they’re at. We customize the data fields to their stipulations and their interpretations of the regulations. And we have it baked in, in our process of vetting the providers and in the provider agreements that we assign to these different biobanks.

Grant Belgard: What mechanisms ensure donor reconsent if the intended research scope changes?

Caspar Barnes: Well, at the moment, we’re not in the process of engaging the research participants. It’s all just focusing on retrospective collection. This problem of, you know, the hundreds of millions of samples out there that are sort of languishing and they’re really expensive to keep and they’re never seen the light of day. The first thing that we can do to help the industry is just to go out to those folks and say, we’ll help you increase this sample exposure and visibility and harmonization of search. We’re going to continue to do that likely for the next six to nine months. But at the same time, what we’re currently doing right now, [?] is building out our own prospective cohorts. And that’s a really exciting pivot that AminoChain folks got, or evolution of the product. That looks slightly different.

Caspar Barnes: That basically involves working closely with clinical sites, closely with advocacy groups, designing custom interfaces and user experiences for patients connected to advocacy groups. And either ourselves sponsoring new collections at those sites or raising money on behalf of these advocacy groups to sponsor collections at those sites. And then when these patients are consented, those specimens will be banked in the specimen center. And the data, the multi- data that’s produced from these studies would be put into a database and access to the database would also be governed by smart contracts. And so if a pharma company would have paid out access to the data for discovery or any other researcher would pay to have access to it for research purposes, in that transaction, we can pay back the people that helped sponsor the collections.

Caspar Barnes: We can pay dividends and royalties to advocacy groups, to trial sites, to patients, to anybody who was involved in the curation of data set. The buyer of that data can use it exclusively for an embargo period. And after which the data would be made available within the decentralized biobank, and it can be repackaged and relicensed in another product to somebody else. And so all this considered within this new prospective collection and data management product that we are soon going to launch, the donors will always have a way of identifying how their data set is used within this decentralized biobank. We use a combination of private-public key photography. They can authenticate how their data set is being used by [?] committed on chain. And from that, they’d also be able to claim rewards if the data is used for commercial in certain ways.

Caspar Barnes: And then through this system, we hope to have a fully incentive-aligned, decentralized, community-owned biobank.

Grant Belgard: If you could only track one KPI for 12 months, what would it be and why?

Caspar Barnes: That’s a good question. So it would still be, I mean, in the context of the specimen center, the one KPI would be requests. It would be, you know, that’s our true north is how many suppliers do we have? How many users do we have? And then ultimately how many requests are we making? And then the context of this new product that we’re launching, sort of the prospective collection metric, the one KPI is licensing data for exclusive use. Like, you know, are we finding people that want to buy access to multi-omic data sets for discovery? And that probably is the most important metric, because I think if we’re able to prove out that flywheel of funneling data, aggregating data, and you know, selling it, then all the other apps on top are easy to build and they benefit from the network.

Grant Belgard: Was there a formative experience that pushed you towards decentralized solutions?

Caspar Barnes: Yes. So not so much like a decentralized solution, per se. It just turned out that crypto was a good way to fix the problem of, you know, biosample tracking and so on. But I certainly did have a formative experience, you know, biosamples in general and, you know, the bioethics of donating tissue. And so very quickly, I’ll give you an overview of that. But I grew up in South Africa, right? You might be able to hear that from my accent. And growing up in Cape Town, South Africa, my mother, she started the charity called Yabongo, which helps women with HIV and AIDS get access to antiretroviral treatments and provides homeschooling support towards kids in the townships outside of the cities. And so growing up, I would spend a lot of time in and out of these townships with my older sister.

Caspar Barnes: And so we had very regular discussions around race, equity, privilege, and especially in post-apartheid South Africa, right? There was always a big emphasis on having these conversations openly and saying, why do we live in this area of town? And why do other people live in this area of town? So trying to find ways to give back throughout your career has always been a really big familial and cultural value that’s definitely now playing into the vision of AminoChain with an aspect of health equity. And the second thing was when I was around 12 years old, I had a malignant melanoma and I was very lucky because it was caught super early. So I just needed one big operation to remove tumor. But since then I’ve been thrown into a world of healthcare, right? And I still remember listening to these doctors explain concepts of healthy cells and malignant cells and how cancer spread and so on.

Caspar Barnes: As a little kid, I just listened with wide-eyed fascination and fell in love with biology at that moment. I was like, I have to work in life sciences in some capacity. And so then since then, I, you know, at the age of 16, would already spend my summers working in [oncolytic?] biovector research, did my undergrad degree in neuroscience, did a graduate degree in biotech, another one in bioethics, all at UCL, Columbia and Harvard Medical School. And so I’ve always been in love with biology since those early days. But the key thing is I can still remember waking up from the surgery and the doctor was standing over the bed and he was holding this biopsy sample. And he was like, check it out. This is what we cut from your lower back. And I was like, dude, that’s so cool. Can I take it home? I want to show my friend. We’re not going to believe this.

Caspar Barnes: And the doctor said, no, we need to keep this biospecimen so we can research it. And I was, of course, too young to understand the connotations of what was going on. But my mom, importantly, was like, yep, sure. That’s for the benefit of science. Let’s sign this consent document and give away the sample. And we never saw it again. And then many, many years later, when I was in the lab at Columbia, I was doing research on somebody else’s donated tissue. And we’re generating all this valuable information, finding all these markers. And I went to my PI and I said, can we tell the patient about this information that we’re generating? And she said, I don’t know where that thing came from. It just came from the biobank. And that blew my mind. I was like, how is that possible? How many people around the world are doing research on biospecimens, generating so much data?

Caspar Barnes: And you’re telling me that none of them know where the samples came from. They all just came from the biobank. And the more I dug into it, it turns out consent rates are incredibly low. There’s almost 25% at some major institutions. The people that are periodically not consenting to having their samples and data used are marginalized communities and patients of color more often than not. And so I got fascinated by the problem and I just got kind of sucked down the rabbit hole where I just did everything I could to try to find an interesting new emerging technology that could fix this. Turns out crypto is great, right? It has all the benefits of immutability and prominence tracking and ownership and agency. And this happened to be around the same time as when crypto was booming in 2020. And so it just seemed like a great match. And I was just like, this is awesome.

Caspar Barnes: Let’s go and try to see what all we can build at the nexus of all these incredible fields, which is emerging tech, life sciences, health equity. And now I’ve just become so mired in this, this like interdisciplinary platform and approach. And so it’s kind of become my life’s work and I don’t think I know. So many consent needs to be disrupted. The current consent model is like as recommended by, you know, the OHRP and the HHS and so on. It’s just like informed consent. That’s it. Here’s one document, write what you want on it, get someone a sign. And then if you see a signature, great. That’s basically your waiver of liability. You know, it’s, it’s not at all a way to meaningfully engage someone in what’s happening with their, their samples or their data and so on. And it’s, that’s kind of like a problem that we find ourselves in right now.

Caspar Barnes: Since 1970s or so, we had a period of, you know, like progressive change within America, right? It was like women’s rights were coming up. There was civil rights coming up and there was all these, you know, bioethical discussions happening as well. We ended up having the Belmont report, 1978, 1979. Of the Belmont report, we came up with these principles for human subjects research, which were, you know, autonomy, justice, and beneficence, right? So of these guiding principles, how could we, you know, have a, a scaffold for involving people in research? Well, informed consent legislation seems to be the best policies. Since the late 1970s or eighties, they were like, okay, let’s try to codify this into law as much as we can or at least make it like public policy that anytime you do research, you can only do it with informed consent.

Caspar Barnes: And it was all done with the, you know, great intentions and it made a lot of sense. And so then since the 1980s, we have to ask everybody for informed consent before they’re involved in a procedure, before they donate the best ones and so on. But, you know, even though those consenting frameworks haven’t changed in the last like 45 years-ish, the world has drastically changed. You know, it doesn’t look the same as it did in the 1980s anymore. Particularly the storage of biospecimens for secondary research has become a booming practice, enormous. In the 1990s, we spent billions of dollars to sequence one human genome. Now we can do it in a matter of days or hours for a few hundred bucks. It’s an enormous progression where we now live in a world where there’s a diaspora of data. And it’s like, all these samples are stored for secondary research use.

Caspar Barnes: And things are becoming increasingly re-identifiable. Things are becoming more and more personal, especially with whole genomes even seen. And with all this considered, informed consent just doesn’t cut it anymore at all. People are asking for a one-time consent document and then they’re doing whatever they want with the tissues afterwards because they’re just getting broad clauses. So all this considered, how do we see a new world where consenting can change the biomedical research industry? Well, we’ve jumped up a new framework called Demonstrated Consent. And under Demonstrated Consent, we can basically take personalized conditions for broad use from research participants. And basically like they’re personalized terms, we take them, we put them as metadata of a specimen, as an NFT. So we have ways to automate the record keeping of the samples.

Caspar Barnes: And then we list them on a platform where anybody can acquire these specimens for research use. Their protocol upholds the consent that the patients originally gave. And if it does, then they can license it, they can use it. At all times, patients have a way to stay informed with the outcomes of research and they can stay informed with how the samples are being used. And so therefore, the blockchain would be demonstrating to you how your sample are being used, as opposed to somebody just saying that they’re using it the way that they will. And that changes the paradigm that actually makes a better experience for the research participants. And you actually have the need for flexibility research, which is like a societal benefit, right? You don’t have to compromise between asks for progressing research and the ask for promoting patient autonomy. And so that’s what we see as the future.

Caspar Barnes: And that’s what we want to embed into the AminoChain protocol. It’s actually like personalized conditions for broad use and an automated way to re-contact and re-engage participants.

Grant Belgard: What surprised you the most in your customer discovery process?

Caspar Barnes: Um, surprises? That’s a good question. Many things are surprising. I think, I think, you know, when starting this out, I thought a biobank was a biobank because, you know, it’s just like, they’re all the same. It’s like, you know, a place where you store samples and that’s it. And I didn’t really realize the complexities that go into biobanking and how many different types of users there are within biobanks and how they are all separate from each other in terms of their, their priorities and their missions and approaches and so on. And so what surprised me, you know, one of the things that surprised me was that you, you have so many different types of doing things in biobanks. Some commercial brokers go and buy remnant material from hospitals and emerging economies. And then they add enormous markups and they sell those specimens to labs in Boston and in San Diego.

Caspar Barnes: And I was like, that’s crazy. I didn’t know that was a practice. And then you try to speak with, you know, other biobanks in America and they’re the part of AMCs, academic medical centers, and they don’t really care about cost recovery at all. What they care about is publications and they care about, you know, insights and they care about all these other things that will make them more eligible for grant funding. And so they’re not brokering tissue. They’re more focused on, you know, moving knowledge forward. And so I thought that was super interesting. Others are independent and they’re part of government labs and others are part of hospital networks. And some biobanks just collect remnant materials from clinical trials, which are associated with the pharma companies. And you’ll never be able to see any of those biobanks.

Caspar Barnes: And so all these things I found really interesting, just landscaping the different customers out there, like speaking with them and hearing what their needs are. It’s been fascinating to have the same conversation with different users, but to see the differences and important factors crop up and motivations for each of them.

Grant Belgard: How did you pitch A16Z crypto differently from life science species?

Caspar Barnes: Yeah, that’s also a good question. So, um, you know, building what we’re building, you have to toe the line between the crypto language and the non crypto language quite delicately. A16Z is fantastic because they have, you know, investors across both verticals. They have a healthcare fund and they have a crypto fund. And so when we were pitching A16Z crypto, we can, you know, pitch the crypto vision and how this becomes the Ethereum of healthcare. You know, the world’s biggest composable blockchain for people to build healthcare Apps. And they get it and it makes sense. And biobanking is the wedge to get there and they love it. But then if you try to say the words that I just said to you there, it’s a, the A16Z bio and health team, they get very confused. And as a matter of fact, that’s like what happened. So we spoke with both of the funds.

Caspar Barnes: And then eventually after a few rounds, we first went through their accelerator program, and then afterwards we’re reinvested as a full portfolio company and so on. Even with an A16Z, we have the practice of pitching both the crypto side and the healthcare side. But all in all, how life sciences VCs look at this as opposed to crypto VCs is, you know, how is this an extension of what’s currently happening today? And if you don’t have to give me complex crypto jargon, but you can just explain in normal language, how already what we see in biobanking lays a precedent for distributed ledger technology to help engage, you know, and to help improve user experiences or improve outcomes or whatever, then it builds a more convincing narrative in their head. So our second biggest investor, Socano is the family office of Paul Allen. They have a lot of life sciences companies in their portfolio.

Caspar Barnes: And so when we pitched them, they were basically our life sciences investor. The language that we had to engage with them was stuff like benefit sharing, stuff like co-ownership of IP, concepts of automating provenance tracking and supply chain management and so on. And if you, if we could just, you know, convey the same technology benefits of the tech that we’re using in non crypto language, then, then eventually it made sense to them. And then it, you know, ticks across all the people that we have on the, on the cat table.

Grant Belgard: What traits do you screen for when hiring at the biology web three interface?

Caspar Barnes: Yeah. It’s a great question as well. The people that are well versed and experienced at exactly the nexus of the two are few and far in between. And so when you find them, you really got to look after them. But then all things being equal, I’d see my job as the founder of AminoChain as being the person to stimulate conversation between the either non life sciences experienced people or the non crypto experienced people, such that they learn about the industry and they become experts at both, or they become at least knowledgeable of, of both the fields in which we’re building. And so we have people just that focused on the life sciences with their PhD backgrounds. They’ve worked in bio sample procurement and, and, and in life sciences research in general.

Caspar Barnes: And then on the other side, we hire people that just have cryptography experience and they just have blockchain engineering experience and they know how to build amazing software. And across both, the main thing that I look for is proactivism. Somebody that just says, just let me take care of that. I’ll, I’ll make sure that gets done. I mean, anybody that is autodidactic, anybody that is, you know, self-starter and proactive, tries to make life easier for their teammates is just an instant green flag. We would sooner have someone, you know, that is very proactive, but maybe less experienced as opposed to someone that’s super experienced, but not very motivated. So across both of those, that’s what we look for. And then second to that, we probably do focus mostly on, you know, the experience and the network that built out within the industry.

Caspar Barnes: So we have some folks that have been doing this 30 years and they’ve got fantastic connections within the space and they can just click their fingers and make things happen. And then I think the last thing as well is people that you can just trust, right? I think that’s the most important thing. So the people that you don’t have to worry if they’re, you know, not working today or if they are working today, just trust that they’ve really bought into the mission and they think that we’re building something incredibly important. And they understand that the faster we built, the faster we could actually help human beings. And so if we have that level of trust across anybody with any level of background and experience, that’s probably the most important thing. And I’m very privileged and like grateful that we’ve managed to build that with the team we have so far.

Grant Belgard: What’s the single best piece of advice you’ve received from a board member?

Caspar Barnes: The single best piece of advice I’ve received from a board member, they give us so many pieces of advice. I think, you know, maybe they sound a little bit cliche, but I think probably the most important thing are the best advice. The only time when you are guaranteed to fail is when you give up or when you stop trying. The whole first year of AminoChain, we were picking pennies, trying to make it work. We were like five people living off of a hundred thousand dollars in New York City, like really trying to make it work. And, and we did, you know, we were super frugal, very resourceful. We incredibly proactive, went out and spoke with everybody and did everything we could to move the needle. We took 250 VC meetings before we got our first yes. And somehow that first yes happened to be Andreessen Horowitz, which was incredible, but it was a long, long, long process.

Caspar Barnes: And the one thing that particular board member I’m thinking of reminded me of the entire time was, well, the only way that it’s a hundred percent not going to work is if you stop trying right now. And that was like a real fuel of motivation that got us through the early days. And that’s, you know, kind of resonate with me for a long time.

Grant Belgard: So looking back five years, what would 2020 Caspar find most surprising about today’s AminoChain?

Caspar Barnes: He would be so mind blown that we found ourselves in the situation that we’re in right now. I think that I, old me would probably be very, I’d like to think he’d be very proud of all the things that we’ve achieved so far, but he also probably been very unsatisfied with how far we’ve come because there’s always more to do. But in 2020, we had the earliest trappings of an idea of AminoChain. And so we knew what it could be, but it was so nebulous at the time. We just knew there was potential. It was not at all clear where we should go. We’ve learned so much throughout the process. I think that old me would probably say, we would probably just be excited for the years to come because it’s like, nothing’s guaranteed. Everything’s difficult. So many people are relying on you. It’s not an easy job at all, but for some reason, you just can’t stop.

Caspar Barnes: And so I think you would be happy that we’ve gotten closer to finding something that’s worked. Honestly, I still think we have a way to go to prove the real product market fit that we need to nail the adoption. But maybe 20 year old me would already thought that we’d taken it further than it could have gone, which means now there’s only one way up to keep going and double down the direction that we’re going in. And so it would be a mix of excitement, maybe pride, but then more so above all else, like motivation to keep going. So I’d like to think that’s what 2020 capital would say where we are now.

Grant Belgard: What early mistake would you warn every tech bio entrepreneur about?

Caspar Barnes: Oh, well, don’t over dilute your capital too soon. And I think everybody says that. And I also see other people warn early stage founders about things around the capital and who to bring on and advisor shares and like over promising equity to people that don’t add any value. It’s not all these mistakes I read about, but I didn’t really know what they meant until I found myself in the situation. And so I guess now I’d pass the same advice on to other early stage founders. Be careful with your capital, do the research into what the term mean, what are drag along shares, what are rights of first refusals, what are all these things. Understand it well, model out what your capital looks like between rounds very carefully. And then if you’re going to give anybody more equity than needed, give it to your team, give it to your employees.

Caspar Barnes: Don’t give it to advisors that are just trying to shop for freebies or investors that are giving you very aggressive jabs. So I would definitely say research and be diligent and careful around how you structure your cap table. And on that note, I’ll just put a short plug for a program I did called VC University through Berkeley Law. It really teaches you the fundamentals of venture capital, which as a founder, very useful to understand the nature of the pressures that your investors are under from their own limited partners and to really have a more holistic understanding of the ecosystem.

Grant Belgard: What vanity metrics do you see startup decks overusing right now?

Caspar Barnes: Good question. Vanity metric. I think the first thing that comes to mind, I’m sure there’s many more, but the first thing that I can think of is like logos. There’s people overhype the logos, right? You know, like I think there’s a team slide and there’s like logos that pop out, but then, you know, there’s, there’s Disney, Amazon, Harvard, and MIT on there. And then you look through it and it actually turns out that, you know, I shopped at Amazon one time and I took an online course at MIT or something, like something ridiculous. And so I think people massively overinflate the use of logos, both on the team side and the customer side, that it can come across as a little bit disingenuous and maybe TAM, Sam, some metrics. I think that slide tends to be really overhyped.

Caspar Barnes: And if people say they have, you know, a trillion dollar addressable market, I always like, you know, focus more on that slide and see what they really mean and what they’re actually building.

Grant Belgard: So to wrap us up, when people talk about AminoChain in 20 years, what do you hope they say?

Caspar Barnes: I hope that they say, wow, look at this case study from Harvard business school on AminoChain. They proved that you can build an incredibly successful business by putting bioethics at the heart of your business model. And this company proved that if you really care about patient engagement, patient experience, and align incentives for human beings that make research possible, then downstream, everybody benefits. You know, it’s not like providing a better consent experience compromises pharmaceutical interests. It actually aligns with bringing drugs to market and helping people. And along the way, you know, it’s a fantastic protocol and it’s crypto enabled and it’s innovative and whatever. But like, I really would love it if people talk about AminoChain as being a company that proved you can make, you know, a lot of success by caring to people first and foremost.

Caspar Barnes: And so that’s, that’s the real mission of what we’re doing. I’ll happily hang out my hat once we, once we prove that out.

Grant Belgard: So where can our listeners go to learn more and how can they follow AminoChain’s journey?

Caspar Barnes: Amazing. Yeah. So our website is just www.aminochain.io. You can check out the specimen center. If you like, you can log on there. It’s totally open, free for anybody. Go and browse through the hundreds of thousands of biospecimens that we’ve aggregated on there. You can also find us on LinkedIn and follow us on Twitter. We’re just AminoChain. And yeah, if you’re a builder in the space on the life sciences side, or you’re a protocol crypto engineer, then please don’t hesitate to reach out to us through our website as well. We’d love to.

Grant Belgard: Caspar, thank you so much for joining us.

Caspar Barnes: Thank you so much for having me, Grant, it’s been a whole lot of fun.

Transcript of Episode 67: Manos Metzakopian

Disclaimer: Transcripts are automated and may contain errors.

Grant Belgard: Welcome to the Bioinformatics CRO Podcast. I’m your host Grant Belgard and today I’m joined by Manos Metzakopian. Today’s episode is special. We’re using this conversation to introduce CellCodex to the world. Full disclosure, I’m a co-founder and the CTO of CellCodex and Manos is co-founder and CEO. We’ll explore what the company is setting out to do, the scientific and engineering choices behind it, and Manos’ path to this point and practical advice for anyone building at the intersection of wet lab and AI. Let’s dive in.

Manos Metzakopian: Wow, this is amazing. Thank you for the invite.

Grant Belgard: So what would you like listeners to know about CellCodex?

Manos Metzakopian: When we started CellCodex, we imagined a world where there’s abundance of drug targets and basically that there is a cure for every disease. And a major development that happened in the recent years was artificial intelligence gaining this capability of taking large sets of data and providing such solutions. That happened with large language models, with ChatGPT, where all text has been collected and you can now interrogate all text that has been around for use and you can gain a lot of speed in your daily tasks. So imagine if you had an AI model for biology, for discovering new drugs. And that model helps you increase drug target discovery efficiency, but also efficiencies going to the clinic and increasing your chances of success once you go to the clinic. Because at the moment, most of the drugs that reach the clinic fail. And there’s a lot of iteration that goes into drug discovery.

Manos Metzakopian: So AI has the potential of solving these problems. Now, for biology, there isn’t this counterpart of data sets that was there for ChatGPT and text. And there is a big need for data so that the right AI models are trained to realize this future. And yeah, and this is why CellCodex has been brought to the forefront as it’s been created. It’s to solve biology’s biggest bottleneck, which is data. And AI, as I said, has the power to transform drug discovery, but it needs the right kind of biological data, systematic, reproducible, and at scale. And that’s what we want to deliver. Our vision is to accelerate the arrival of the world where every disease is curable. And the first step is giving model builders, AI model builders, and drug target hunters the right fuel, which is the data.

Grant Belgard: So CellCodex is a CRO that generates AI-ready perturbation data at scale.

Manos Metzakopian: That’s correct.

Grant Belgard: So what problem in biology or drug discovery feels most urgent to address right now, and why start there?

Manos Metzakopian: So at the moment, because of the arrival of AI models that can solve these big problems, the creation of superior AI models is moving at a very fast pace, almost at the pace of weeks and months. Whereas a data generation that can feed these models and allow them to be trained, it’s still very slow. And it’s moving at a speed that is not satisfactorily reaching the speed of model creation and testing. So the most urgent gap is reproducible perturbation data we have. And we have plenty of observational data at the moment. However, these are snapshots of what cells look like. So from observational data in biology, we have almost 14 times the amount of data that was there to train ChatGPT. However, it’s the quality of the data and the kind of data that is available that is important.

Manos Metzakopian: And unfortunately, we do not have that right type of data, the perturbation data, the intervening data in cell identity, cell state, and cell function. Without that, AI can’t move from correlation to causation. We started there at CellCodex to create large-scale perturbation data to solve this problem and to allow AI, artificial intelligence, to realize its promise, speed up drug discovery.

Grant Belgard: When you imagine the ideal outcome of this effort five years from now, what does success look like to the end user?

Manos Metzakopian: The success rate for the success is very simple for the end user. It looks like there’s faster drug discovery programs, fewer dead ends, more success in drug target identification, and higher success in the clinic. And for those to be powered by our AI enabling data sets. That’s how I see our success in five years from now.

Grant Belgard: What kinds of decisions do you hope our work helps people make faster or confidently?

Manos Metzakopian: There is a lot of work that goes into drug discovery that takes many, many years. And we can speed it up at the rate of weeks and months to be able to make these decisions in weeks and months versus years. And that includes which targets to pursue, which mechanisms are causal, which disease models are worth investing in. Right now, those decisions often take years and huge budgets, and we want to make them faster, cheaper, and with higher confidence.

Grant Belgard: What milestones are you most comfortable sharing at this stage, and what should listeners watch for next?

Manos Metzakopian: So a major milestone for us is that we’ve set up and are continuing to build our platform, the CellCodex platform at the Babraham Research Campus at the moment, where we are going to launch our first collaborations, partner projects, client projects. We also want to publish benchmarking data sets that show what AI-grade data really looks like. So listeners should watch for collaborations where our data sets are powering new models or enabling novel drug targets and emerging new drug targets due to our data sets and enabling of client models, our AI models.

Grant Belgard: When building virtual models of cellular behavior, what principles guide how you define the unit of prediction or simulation?

Manos Metzakopian: So we think of units not as just a number of cells that are being evaluated as it’s being done in observational data, but we are also thinking of cell states, cell states under perturbation. So a meaningful unit for us isn’t just a cell at rest or it’s just in its normal environment. It’s a cell that is responding to a defined change. This is the building block for causal AI modeling, I would say.

Grant Belgard: Where do you draw the line between a correlational model that’s useful and a model that supports causal reasoning?

Manos Metzakopian: So a correlation model is useful for pattern recognition, but causation comes when you’ve systematically perturbed the system. So our role here is to generate that causal data so customers can build models that go beyond what co-occurs and moves to what actually drives change. For example, in disease, point mutations can lead to changes in cell state, and these are not just co-occurring mutations. They are driving the change. So we are interested in data sets that empower models that can quickly identify mechanisms that actually drive change in cells.

Grant Belgard: In your view, what types of measurements provide the most leverage for learning cell state transitions under perturbation?

Manos Metzakopian: For us, at the moment, we need single-cell multi-omics, and we have two major capabilities to sequence RNA, cells messenger RNA at scale, but also to acquire epigenetic changes, the epigenetic landscape in the cell through ETAC sequencing. So that captures which areas of the genome are open, and so you know which genes are expressed, but also correlate those to which areas of the genome are open as well. So these two data sets provide, number one, which genes are expressed, how are they changed under perturbation, and very importantly, which features of the epigenome change. So when you sequence, when you have ATAC sequencing, you can also correlate the changes in many features of the genome to the gene expression changes as well. So that adds a lot more information to interpret causation versus correlation.

Grant Belgard: How do you think about biological context, cell type, state, microenvironment, when designing a modeling target?

Manos Metzakopian: That’s a very, very important question. This is essence of what we do in CellCodex. So in CellCodex, we have the functional genomics capabilities to produce these large-scale perturbation data sets through our genetic screening approaches and gene editing technologies. However, the foundation that can lead to the right type of data are the models that we would use to generate these data sets. And so we design our experiments according, of course, to what the clients would need with the cell identity and function in mind, developmental states, co-cultures, which cells need to be together in the dish, and the microenvironment that they are supposed to be growing in. So you take all of that together, and then you have your human cellular models that you would use for your perturbation experiments.

Manos Metzakopian: And if you want to think about it, a perturbation in a neuron means something very different than a perturbation in a fibroblast. So that’s cell identity. So we co-design with customers to choose the context that matters to their question and to problems that their AI models would want to tackle.

Grant Belgard: What would you count as a falsifying result that sends you back to the drawing board?

Manos Metzakopian: A very important thing is the quality of the data, and a lot of it goes into data reproducibility. So we put measures, quality control measures, in place at every step of our platform, so our data sets are reproducible across batches. It would be very challenging if we don’t have batch-to-batch reproducibility. If you think about the cellular models that we are using, so every time we perform tissue culturing and using the cellular models to produce the data, they need to be the same and reproducible. And the data sets that are coming out of these models, the perturbation data sets, need to be reproducible. So we have very strict metrics around that. I would say that would be one of the major falsifying results that can happen in the platform. And we have very stringent mitigation strategies for that.

Grant Belgard: And when you plan data for model training, what are the first three design decisions you lock in and why those?

Manos Metzakopian: Most important thing is the context of which cell models to use, because that’s, if you think about disease, they don’t happen in isolation. They happen in a specific context with specific cell types involved and cell-cell communications happening there. So the cell models to use are one of the first decisions we need to make. Of course, they need to be applicable to our screening strategies as well. And then which perturbations to apply? Is it a gain-of-function perturbation, like using CRISPR activation, or is it a knockdown perturbation, or are we looking at completely knocking out the gene? So which perturbations, and depending on the experiments, the different scales. So we might need a few million cells for an experiment, or hundreds of millions of cells for an experiment.

Manos Metzakopian: So if you’re thinking of foundation models, for example, versus very specifically trained models that would need fewer cells. And which readouts, right? If you’re thinking of omics readouts, which of those readouts, so that you can balance resolution, cost, and downstream utility.

Grant Belgard: What’s your approach to quality control from sample prep through to process matrices?

Manos Metzakopian: Our approach is to have quality control steps embedded in every part of our platform and our process. And to have the right type of standard samples or tests in every component of our measurement. So that then we can always have a good measure of the quality of the data that’s coming through. So from tissue culturing and the cells, quality of the cells that we are using for our perturbations, the quality of the material that we are extracting from the cells, and finally, the quality of data that is being extracted from our cellular models. I would say we have very clearly defined pass-fail criteria up front for customers to know what they’re going to be getting regarding quality of experiments and data.

Grant Belgared: What’s your stance on foundation-style pre-training versus task-specific architectures?

Manos Metzakopian: So I think both are going to be important. You will have customers that are looking for models that can generalize very broadly. So they would be building foundation models, and those would require vast, diverse data sets. So there’s going to be breadth and depth required for such models. And task-specific models will need more precisely curated data sets coming from very specific contexts. And in both cases, that will decide the number of cell types and complex cultures that we would be using to deliver the data sets for both types. Foundation models will have quite broad utility, but the task-oriented ones would be more specific. And we will be producing data sets for both types of model training approaches.

Grant Belgard: What does a convincing benchmark look like to you for the model understands a cell response?

Manos Metzakopian: It can be covered by just one word, I would say, replication and validation. So if, sorry, two words, replication and validation. And that is that we are able to reproduce the same perturbations providing the same data. So that would be replication. And also validation, the outputs of the models that can be validated in turn. So I think these are going to be very important benchmarking tools that we have. That’s how I would think of it in simplistic terms.

Grant Belgard: How do you separate evaluation of biological plausibility from pure predictive accuracy?

Manos Metzakopian: So I would say that it’s very important to focus on biological plausibility. Because if the data itself isn’t biologically valid, accuracy metric will matter in the sense. So it has to be applicable to the scientific challenge that the client wants to tackle. So I would put a lot of focus on biological plausibility, initially, especially in scientific design, in the experimental design.

Grant Belgard: What forms of external validation replication blinded test challenges feel most meaningful?

Manos Metzakopian: It would be great if independent labs can replicate. If you think of it from a replication point of view, if different labs can generate the same data with the same approach, that provides a lot of confidence. But in our case, I think we would think of it as customers successfully using our data to build their models and generalize to new biology. So if they are able to use our data, generalize to new biology, and identify targets, solve their biological problems, and expedite the therapeutic discovery path and increase its effectiveness, then I would say that’s the most meaningful external validation.

Grant Belgard: Who stands to benefit first from this work, and how might they plug it into existing workflows?

Manos Metzakopian: I think at the moment, there is a race happening of different entities and institutions and consortiums and consortia that are working towards delivering a model that can solve a lot of the drug discovery problems. And that includes biopharma teams, and that includes biopharma teams, and biopharma teams, and consortia, and so on. But I think what is currently being understood that it’s not going to be a one-dataset-fits-all. It’s going to be models that are going to be trained to solve specific problems, and they’re going to be requiring their bespoke data sets to be trained with. And so I think it’s going to be less of a race towards the best model, but more of a joint effort to generate the right data for the right models and solve pressing issues in the world. And I think that that day is upon us, for sure.

Grant Belgard: What kinds of collaborations or partnerships would be most impactful at this stage?

Manos Metzakopian: Companies that have bottlenecks in their pipelines where our data can actually resolve that issue.

Grant Belgard: How do you weigh openness, sharing resources, or benchmarks against the need to build a durable business?

Manos Metzakopian: I would say it’s very important to make sure that we are leading in the space of high-quality data sets, AI-grade data sets. And we should think of best ways of sharing benchmarks and best practices openly. However, the large-scale perturbation data sets are contract-delivered, and so there needs to be a balance that ensures both impact and sustainability.

Grant Belgard: What drew you personally to this specific problem space?

Manos Metzakopian: I have always been involved in projects and challenges that require large data and perturbation data. Most recently, we’ve used this know-how in the cell programming field. So to democratize cell types for drug discovery research and cell therapies, and that never required large-scale data sets and so on. And during my time solving these problems in academia and in industry, I realized that the potential for AI to solve the drug discovery bottleneck and lead to a world where there are cures for every disease requires us to rethink the way that we produce data, the quality of the data, its reproducibility and its scale, and the context at which it is delivered.

Manos Metzakopian: And so as I was progressing in my academic and industry career, I’ve realized that setting up a platform like this, which is CellCodex in this case, to generate AI-grade data is timely and very, very important to do so now, where we are at the verge of arising to artificial intelligence-enabled solutions in therapeutics.

Grant Belgard: Looking back, what set of experiences most shaped how you approach leading a science-driven company?

Manos Metzakopian: The most important experience that I had during my academic career and my industry career is managing people effectively, making sure that we are all goal-driven, we are ambitious, and we are enjoying what we’re doing. And in my academic career, I’ve mentored PhD students, master’s students, and postdocs, research assistants, and technicians. It led to amazing work where we’ve published over 30 scientific manuscripts in the fields of genetic screening, cell engineering, and drug target discovery. And similarly, in industry, leading larger teams, the most important thing that leads to success is the team, the people that are involved in driving the work and the goals that we set ahead of us. So I think goal-setting and the people that are along for the journey are the most important pieces of the puzzle.

Grant Belgard: How do you structure your day to balance science, product, people, and operations?

Manos Metzakopian: It’s not always easy to balance between everything. It depends on the stage at which the activities are. If it’s joining a mature corporation where they’ve already set off and they’re on a journey, or in this case, CellCodex, where we are just launching, everyone in CellCodex wears multiple hats, and we try to support each other and help each other so that we can deliver the needs of the company. And I structure my day where I look at the needs of the people, if there’s any way I can help in their day-to-day activities, the needs of the company, and in designing the strategy, and what type of products we’re going to have, and offerings. And of course, now, when launching, we’re thinking of operations. How are we going to operate most effectively? And I would say, at the moment, it’s split 30% equally throughout everything.

Manos Metzakopian: So I would say it’s equally divided across strategy, products, and operations.

Grant Belgard: What advice would you offer to scientists considering a leap into company building?

Manos Metzakopian: You’re not going to feel ready. So at any time point, especially when it’s your first venture. So I would say, if you have the right ideas, and you have a very strong feeling and passion about these ideas, you have people equally passionate with you, and you can work together to make them materialize, then I would jump in, and I wouldn’t wait until you feel fully ready. You probably won’t get to that type of feeling. And it’s not a bad thing. And bottlenecks are not going to fix themselves. So if you see one clearly, then that’s your opportunity to jump in with your ideas to solving a problem in the world.

Grant Belgard: What practices help a small team avoid cargo cult, ML, or overfitting ideas type cycles?

Manos Metzakopian: I wouldn’t chase hype cycles. I would ask if the method helps explain or actually lead to a solution. So I would really think and investigate very, very, very well, very deeply, if a new direction, a new tool, a new approach is really going to make a big difference. And ask yourself if it’s worth the investment. So I wouldn’t chase. I would investigate and research what new things come out.

Grant Belgard: What advances outside your control would most accelerate your roadmap?

Manos Metzakopian: So that’s a great question. So outside, so currently, as I’ve said before, throughout this conversation, this podcast, there are a lot of companies out there that are generating their own artificial intelligence models, and they are using them for predictions that can progress drug discovery. Now, there are a lot of companies that are doing that at the moment already, and there is a big need for data. However, as soon as these models start showing the power that they have in increasing drug target discovery and driving efficiencies in therapies, there’s going to be even a larger need, and there’s going to be a larger number of models that are going to be generated to be trained, and there’s going to be a lot more data that’s going to be needed to train these models.

Manos Metzakopian: So I would say that since there’s going to be such a huge need for data advances that can increase the number of cells that we can analyze in a multi-omics context and technology development that can allow us to analyze multiple modalities from similar samples, all of these will allow for better data, larger-scale data that can provide the fuel that these new models will need in the future.

Grant Belgard: Well, Manos, thank you for sharing the CellCodex vision and the thinking behind it. It was nice having you on today.

Manos Metzakopian: Thank you very much for the invitation. It was a great, great conversation. Thank you. For listeners who want to follow along, the best place is cellcodex.bio and also our LinkedIn page. If you enjoyed this, please subscribe and share with a colleague who cares about building predictive biology. Thanks.

Transcript of Episode 66: Eva-Maria Hempe

Disclaimer: Transcripts may contain errors.

Grant Belgard: Welcome to The Bioinformatics CRO podcast. I’m your host, Grant Belgard. Today, we’re joined by Dr. Eva-Maria Hempe, who leads NVIDIA’s healthcare and life sciences business across Europe, the Middle East, and Africa. Eva-Maria, trained as a physicist, earned a Bill and Melinda Gates funded PhD in healthcare service design at Cambridge, and has since moved through roles at the NHS, Bain & Company, VMware, and the World Economic Forum before joining NVIDIA. She now guides strategy for applying accelerated computing and generative AI, think BioNeMo, Parabricks, and DGX Cloud, to genomics, drug discovery, medical imaging, and more. Eva-Maria, welcome to the show.

Eva-Maria Hempe: Hey, great to be here.

Grant Belgard: So what do you do day-to-day at NVIDIA?

Eva-Maria Hempe: I think in general, my day-to-day oscillates between two major poles, like working in the business and working on the business, or playing the short game and the long game. So on the one hand side, I am responsible for the business. And so that means we have to deliver revenue because if you don’t deliver revenue, you’re not a business, you’re a hobby. And when, on the one hand side, I have to hit a revenue number because if you don’t have a revenue, then you’re not a business. But on the other hand, NVIDIA is all about the long game. Like we are creating markets. We are building things that haven’t been built before. And so it’s really about striking this balance. And what it means, very practical, is on the one hand side, as I said, working in the business. So I have customer meetings.

Eva-Maria Hempe: I work with my team. We’re discussing strategies and tactics, like what should be our sales place? How are we going to work with startups? How are we going to work with this customer? I check out KPI if I see like, are we on track to delivering the revenues that is expected of us? I do a lot of talks and evangelizing to spread the message that NVIDIA is so much more than just GPUs that we have all this great software out there as well, which is super helpful and super valuable to our ecosystem that people can save a lot of time by building on top of what we put out there. So that’s the operational part. And then there is the working on the business. So really the more strategizing, making decisions on, should we focus on enterprises or startups? Where within healthcare should we focus?

Eva-Maria Hempe: To whom do we talk about which kind of topics? To which degree are we focusing on the sale? But where do we see new areas emerging which maybe aren’t driving a sale or even a lot of compute initially, but where we really believe that there are, A, making an impact. And then if they make an impact, eventually it will turn into revenue, which is one of the real beauties about working at NVIDIA that the company is set up in this way to build, to disrupt, to change and to, yeah, you have this luxury almost like it’s a bit crazy to call it luxury, but in a lot of businesses, it’s a luxury you don’t have to really work on your business than just working in the business.

Grant Belgard: So BioNeMo just went open source. Can you tell us about that and what pain point it solves?

Eva-Maria Hempe: Yeah, so in general, as I said before, we’re trying to do at NVIDIA, we’re trying to lift up the field. So we’re not looking for the quick buck. So that’s why we’re not looking to, we’re not gonna change the field by collecting licensed revenues on BioNeMo, but we think BioNeMo is a super interesting, super valuable tool for the community. And by putting it out there as open source, we can just make it much more available to a lot more people. And also we can increase the number of people who are contributing to it with their ideas and making it into something that is a lot more valuable to the community and more powerful and much more in line with the community. I think around the same time that we made it open source, we actually also, we changed it.

Eva-Maria Hempe: Like we turned it into, it has two pieces these days, the one is BioNeMo Framework and the other one is NIMs. So Framework is really, it’s also a collection of microservices, but it’s a collection of microservices, which you need to train and deploy models. So it has a curator and an evaluator and a guard railing part to it. And you can use all of these, you can use any of these, whatever helps you to put out models in a better way. And then we have NIMs and so NVIDIA Inference Microservices and some of them are biology specific. So we have some on folding, we’ve got some on generation, we’ve got some on docking, and you can put this together into reference workflows, which we call blueprints.

Eva-Maria Hempe: I often say it’s a bit like, if you think of a big box of Legos, it’s like the building plan, how you build the most basic thing out of them and then you can play with it and turn it into all sorts of other things. But in general, what we’re trying to do with BioNeMo is really solving the main pain points of drug discovery. So drug discovery is slow, it’s expensive and then also quite technically challenging if you want to use computer aided drug discovery. And so here we’re giving researchers tools to handle complex data, to collaborate and just in general, we wanna have an advanced biomolecular research framework out there that people can use and that they can do their best work with.

Grant Belgard: And for our listeners who aren’t already familiar with BioNeMo, can you give a quick primer on what they can do with it?

Eva-Maria Hempe: So, as I said, it is mostly about computer aided drug discovery. So one way I usually explain it, we have another framework called NeMo and that’s not by coincidence. So NeMo is all about training, deploying models that have to do with language, but by now it’s actually also multimodal and BioNeMo is that for the language of biology. So if you think about a sentence has like words and observes grammar and the same way like a molecule has atoms and observes the laws of physics and chemistry. And so that’s a bit the analogy there. And so the same way that with our language model, you might have proprietary data and you might wanna train a model on this or you might wanna fine tune a model with new data, you can do the same thing with biological data.

Eva-Maria Hempe: If you have data coming in, you can curate it and then you can also make sure, so that’s the curator part, then you can also evaluate it against certain benchmarks. So how good is my model? And then finally you can also make sure it has certain guardrails, so it doesn’t do certain things that you don’t want it to do. And so that’s, yeah, that’s in a nutshell about it. It’s about training, deploying and serving biological models for drug discovery.

Grant Belgard: So AlphaFold has made a huge splash in the structural biology world. What do you think is the next big thing that would be GPU enabled in biology?

Eva-Maria Hempe: For me, AlphaFold is really like, I’m a physicist. So I know when I did my PhD, which in my mind hasn’t been that long ago, we locked up PhD students for three years in a basement to find out the 3D structure of a protein. And now you can just do it on a computer. You can go to build.nvidia.com where we host the NIMs, I said before, and we have a model there and you could fold a protein in like a second live on your computer. And it’s just mind blowing. It even works on my phone. I’ve done it during presentations on my phone. So I’ve folded a protein on my phone within less than a second. In general, there are certain things around AlphaFold. There are certain gaps. So it has problems with dynamics. It has problems with multiple conformations. It can’t do disordered proteins.

Eva-Maria Hempe: And 60% of human proteins have at least one intrinsically disordered region. It’s also not great with protein ligand and nucleic acid interaction. So there are a whole lot of things which it cannot do. And so these are actually also the things we see in the field where a lot of work is going on. And as NVIDIA, we’re doing some research ourselves in the spirit I said before, in trying to lift up the field and trying to show what’s possible and trying to also inspire other people to go further down that path. And so we’re doing some research ourselves. We’re doing a lot of research in collaboration with all sorts of other people. Sometimes we’re open about this. Sometimes it’s not disclosed, but yeah, we’re seeing a lot of things that are going on.

Eva-Maria Hempe: And what we’re seeing in particular in terms of frontiers, I would say, are four things. So we see how do you deal with larger complexes and assemblies? How do you deal with post-translational modifications? How do you deal with dynamics, molecular dynamics? And then also how do you deal with protein design? Like how can you turn AlphaFold around? Like with AlphaFold, you have the sequence and you want to know the 3D structure. Can you have a 3D structure and figure out what is the sequence behind it? So there’s a bunch of work going on in the space and I think it’s going to be super exciting to see what will come out of that.

Grant Belgard: How do you see DGX Cloud changing the barrier to entry for academic labs?

Eva-Maria Hempe: DGX Cloud is like an interesting way, which is part of what we offer. And maybe it’s easier to understand in the greater context of what we offer. So in general, we are very much agnostic of what GPU you’re running your workloads on or what NVIDIA GPU you’re running your workloads on. And that is a huge advantage for people who are working with our software because we don’t want to lock anybody in. The only commitment you’re making is you’re going to work on GPUs, which I think is not a bad lock-in. You’re not locked in any other way, but that you’re going to be using GPUs. And those GPUs, the answer what GPUs are the right ones for you will again very much depend on your situation. Like, do you have a data center? Is your data center big enough? Has it liquid cooling?

Eva-Maria Hempe: Does it have enough electricity? Do you even want to run a data center? Or do you have big spikes where you need really high performance computing capacity in a short amount of time? And DGX Cloud is following our reference architecture. So it’s really all the different components, the GPU, CPU, networking perfectly aligned with each other. And it’s in the cloud, it’s on demand. So what we see it used quite often for is spike. And if an academic lab has that, if a lab is trying to train a huge model, it can be the right thing for the lab. And it could be a great way as well to showcase the power of it, but it’s not always the right solution. Sometimes it’s also worthwhile to build your own on-prem capacity or to go with more conventional cloud capacity.

Eva-Maria Hempe: So I think it’s an element of a larger compute discussion, but it definitely allows academic labs if they have the funding, if it’s basically baked into the grants to really get top-notch performant GPU computing on really short timescales.

Grant Belgard: And at what stages in the process does AI assist drug discovery today?

Eva-Maria Hempe: Pretty much along all of them, I think we see different levels of activity. So we see a lot of really early discoveries. So it starts with things like finding new targets, which I think is an interesting one. I think it’s one where we don’t see, I think you could see even, I would hope for even more activity. Somebody told me the other day how many people are working, how big the overlap is between working on the same targets. It’s mind blowing. And for example, what we talked before, intrinsically disordered proteins is a super interesting area to really find new targets, to be able to address parts or proteins, which so far have been undruggable.

Eva-Maria Hempe: And we’re working with a company there, they’re called Peptone, and they actually, AI supported, have found a method to figure out the structures of disordered proteins. So I think this was super exciting. So we’re starting there. And then of course, we have all the virtual screening workflows in terms of, okay, you have a target, you fold the target. Then you have something like MolMIM or like a generative model, which starting from a particular small molecule creates all sorts of variations of that small molecule. And then you take your protein and your multiple variations of small molecules you generated, and then you use another AI model, which can calculate how well they fit together. And as I said, that’s an area of active research as well.

Eva-Maria Hempe: How well can you really calculate those bindings? And again, another company we’ve worked with, they’re called Inoform. They can actually also do a, they can create models that fit into a particular, or molecules that fit into a particular cavity. So there’s a lot of interesting things around there on the real fundamental level. But then there’s even more to it. There’s, we’re trying to figure out how can we also, or companies are figuring out how can you apply AI to pre-clinic?

Eva-Maria Hempe: And then even in clinical research, or the clinical stages of drug discovery and drug development, there is still so much that can be done because so many drugs don’t necessarily fail because the biological mechanism isn’t there, but often also because you can’t recruit patients, you can’t recruit the right patient. And again, AI can actually have a huge contribution to solving these kinds of problems. And then you can go into manufacturing and selling drugs. So I always tell my clients that AI is a topic along the entire value chain. And we are seeing applications today along the entire value chain. Like every single step, there is somebody working on something and a lot of progress is being made.

Eva-Maria Hempe: You still have the whole issue that just things take a very long time because like clinical studies just take the amount of time they take. You can have a bit of time out there by doing optimized recruiting of trial participants, which is usually a pretty of a delaying factor, or you can use AI also to speed up the data analysis and regulatory writing, clinical writing, submissions processes. So there is some speed up you can do there. But I think in terms of the speed up is more happening in the earlier phases of drug discovery. And then in development, we really have more of a trying to figure out where do they work. So a lot of work I see in that area as well is around biomarkers.

Eva-Maria Hempe: Again, figuring out what works for which patients so that it feeds back into the early stages, but then also once you’re in trials, you have the right patients in your trials and you have a better chance of actually making it through phase three, doing efficacy. I said about all those different ways, how AI can help with the preclinical part. And there is actually real good data on that by now. So, and SILCO is really famous about this and they were smashing it. They had 22 developmental candidates between 2021 and 2024. And actually they were able to get on average to a developmental candidate within 13 months. So around 70 molecules synthesized per program. And the fastest was like nine months and the longest was 18 months.

Eva-Maria Hempe: And this is just like a huge, huge speed up to what you usually see, but these kinds of processes take years. Interesting, so that’s the preclinical phase where it’s really about the speed up and you can also go from target and lead identification over lead optimization in 46 days these days. So all of this is amazing. And I said before in the clinical studies, it’s then really about being better. And there was a paper which came out last year where they looked at AI discovered drugs. And for phase one, the success, probability of success was twice as high as for regular drugs. And it was still pretty bad, but it was twice as high. And then for phase two, it was in line with the averages, but for phase two, the numbers started to become quite small.

Eva-Maria Hempe: And for phase three, there wasn’t enough data. But if we assume this holds, if you assume you’re twice as successful in phase one, which is not unrealistic because phase one is all about safety and with better models, we get better idea of target effect, and then phase two and three about efficacy and a dosing on part, then this actually means we’re going from one in 10 drugs, making it to markets to two in 10 drugs. It’s still a lot, but it’s basically, it’s halving our cost per drug. And if a drug costs these days, on average $2 billion to make it to market, saving a billion dollars per drug. So this is huge. Your potential is huge, which I think is why we’re all still working on this despite all the problems we talked about of long timelines and difficulties to get funding.

Grant Belgard: Where are the biggest talent gaps in bio AI today?

Eva-Maria Hempe: I think it’s really about speaking multiple languages. And the question is also talent where? So we have and– and what keeps things from reaching or from reaching impact. So I think if you look at a lot of the biotech, tech bio, we still have the issue that the entire pharma ecosystem is set up in a particular way. Somebody said it the way, like it’s a coin flip. And we know that the coin is unfair. We know that heads gonna come with a 10% probability. Now what these companies are doing, they’re actually trying to improve the coin minting process. So by using AI, we’re trying to mint better coins. We’re trying to mint a coin, which has a 20% chance of heading up, landing heads up. But this is really hard to prove.

Eva-Maria Hempe: And the entire system, the people in the VCs, all their mindset is like a biotech investor mindset. And they’re looking for the things around a 10% coin flip probability. And it’s really hard to evaluate this. Is this really going to get us this lift up or not? And different to other areas of AI like quant trading where you have immediate feedback, you change something, okay, you’re gonna make more money. Great, let’s do more of this. Here, it’s almost the complete opposite of quant trading. You have like 10 years until you see whether it works or not. And I think that’s actually one of the biggest gaps.

Grant Belgard: Even with the 10 years, it’s small in, right? So it trickles through after 10 years.

Eva-Maria Hempe: And so, yes, I think we need to have more people who speak multiple languages of AI and of data science and of biology. But I think we’re starting to see some of that. But I think it’s really more the system as a whole and the incentives and the structures and just the fact that we’re dealing with biology, which takes 10 years to come. But I’m still optimistic.

Grant Belgard: What are your thoughts on community standards such as OpenFold and so on? Are there areas where there are glaringly obvious missing standards or areas that you think are still being held back by a lack of standards?

Eva-Maria Hempe: At NVIDIA, we are big believers in open source. So we think it’s the one way to really harness the power of community. And we are big believers in the community. NVIDIA is all about communities, about ecosystems and us doing our part to help the ecosystem develop, which is why so much of our software is actually open source because we believe in the power of this approach. And we really wanna support it to come to full fruition.

Grant Belgard: Well, it’s essential to save biotech and pharma, right? The internal rate of return on R&D has been abysmal below the cost of capital for many years now. And at last that turns up.

Eva-Maria Hempe: It’s actually interesting because of those $2 billion per drug or one and a half billion dollars per drug, only I think it’s around 300 or so are the actual cost. All the rest is the cost of the failed drugs and the cost of capital because the capital is just locked up for such a long time and you have so many failures all around. And the other thing I think, I don’t know, you’ve probably seen it, it’s called Eroom’s Law. If you take how many drugs $1 billion in research spending buys you, it’s a logarithmic downward over the last 70 years. This is not recent. This has been going on forever, but it’s just starting to get into areas where it’s just really, you just can’t continue this way. We just need a different way of doing things.

Eva-Maria Hempe: We just can’t continue spending more and more and more and getting less and less and less.

Grant Belgard: So shifting gears, let’s talk about your own journey. What pulled you from physics to health?

Eva-Maria Hempe: It was the impact. So I was sitting there in my lab. So I was doing quantum optic, which means I’m sitting in a dark lab because I was dealing with optics and lasers. So you don’t want daylight messing up your experiments. So you go in in the morning, it’s dark. You leave in the evening, it’s dark. And during the day, it’s dark. And I was just thinking to myself, what is this going to do for the world? And back then we kept saying, oh yeah, this could be used for quantum computing. But back then I was like, well, but this is going to be at least 15 years until anything useful. And I have to say, this has been more than 15 years ago by now. So I was just like, okay, is this really it? But then as with those decisions, usually two things have to come together.

Eva-Maria Hempe: And the other part, which was for the ignition to really change tack was just meeting the right person at the right time. So I met this girl and she was an electrical engineer by training. And she studied how procurement processes at the hospital affect patient safety from with this very scientific engineering frame of mind. And I just thought that it was fascinating. Like all the way I’ve been trained to think, which like I really liked the scientific method. I really liked this way of thinking, but applying it to real world problems. And that’s how I got to study healthcare service design.

Grant Belgard: Are there any insights from your PhD that you still use?

Eva-Maria Hempe: Yeah, I think it’s really that organizations are an interplay between structure and people. And that sounds very simple and very obvious, but if you’re designing an organization, you’re not actually designing an organization. You’re designing almost a scaffolding for the organization to grow around. You’re giving some structure, but an organization isn’t the org chart. It isn’t the policies. It isn’t the trainings. It’s the people which are populating those structures, which are interacting, which are meeting each other or not meeting each other. And I think that was a really important insight which has like, it pops up everywhere. Now, one of my big challenges at work is like how do I get enterprises to adopt AI?

Eva-Maria Hempe: That’s again, an organizational question. As much as a technological question, actually technological question is like, maybe not even half of it. A lot is really about how do you get people to adopt it? How I get people to use it? What are the incentives they’re listening to? Who has power in this organization? How is this organization really structured? So yeah, I still use some of the things I learned, I studied.

Grant Belgard: And what did you learn in your time with the NHS that you think tech sector often misses?

Eva-Maria Hempe: I think in the tech sector, it’s easy to look at everything through a technological lens that, oh yeah, we can improve this, we can do this. But a lot of my research and my work was about design thinking, which is very much empathy. You start with the end user, you immerse yourself into the end user. Ideally you get to observe, you get to shadow, but you get a real idea of what are people doing and what’s the real problems and how can technology help that? I think this empathy, this user-centric view is sometimes a little bit missing in tech. I think what we also discussed before, you’re creating a great tool and maybe the people you tested it with like it, but it has to fit into the workflow. It has to fit into the real life. It’s all about minimizing friction.

Eva-Maria Hempe: I was saying the other day, just like if you wanna drive real value in organization, it’s about having something that has as little a friction as possible and as much immediate value as possible. And then you’re gonna see adoption. If it’s high friction, it has to have even higher value. If it’s low value, it has to have even lower friction, but ideally it has both.

Grant Belgard: Can you tell us about your time at the World Economic Forum and how that impacted the work you do today?

Eva-Maria Hempe: Yeah, the forum really is about multi-stakeholder and what role policy plays. And again, about what are the right incentives and how can you align the incentives of multiple different parties towards a common goal. So what I did there, it was about the future of healthy. So how do you make staying healthy a business versus having people get sick first and then making them healthy again? I mean, that’s an established business model, but why are we there? Why can’t we just keep people healthy in the first place? And there it’s really about thinking through the food industry. How can we make it a better business for the food industry to sell healthy food? How can we make it better for the doctors to be paid to keep the patients healthy?

Eva-Maria Hempe: There’s models for that where they get basically paid per patient in their catchment area, but they don’t get paid for the procedures they do, but they get like a fixed fee. It has all its pros and cons, but really think through things from a joint value and joint incentive point of view. And like I said, again, when you’re trying to change big systems, whether it is an organization or whether here it is like a multi-organizational system, it’s really important. And this is something I think I couldn’t imagine a better place to learn how you navigate these things, how you deal with politicians, how you deal with all the different lobbyists and all the different interest groups and really try to drive towards a common goal. And I think there’s no better place than the forum to learn that.

Grant Belgard: Can you tell us about your time rowing in Cambridge and did that develop you in any way that’s useful today?

Eva-Maria Hempe: Yeah, I got to Cambridge twice. The first time I went to Cambridge, it was for a summer research as part of my master’s thesis. And I knew people and they made some connections for me. And so I was at Cambridge during the summer before the freshers arrived. And then the freshers, so the first year students all came in and all the clubs started recruiting and the rowing club started recruiting and they tried to recruit me. And I was like, yeah, no, I’m only here for a few more months it doesn’t make sense, I should still do it. And I didn’t do it. And then I came back to Germany where I was finishing my studies and everybody was like, oh, you were in Cambridge, did you row? I’m like, no. And then I really regretted it. I was like, well, I really should have.

Eva-Maria Hempe: So I promised myself if I make it back in for my PhD I’ll give rowing a go. And so I did, and initially I wasn’t that good. So I was in the second novice boat. I didn’t even make the first novice boat. I was in the second boat, but then I just kept at it. And I barely made the first boat in the next term. There’s three terms in Cambridge. And then in the third term, I was still in the first boat of my college, of my part of the university I was at. And then I was around for the summer. So I thought, okay, the university team is doing a summer program. I might as well try that. So I did that. And then they try to funnel you into joining the team full time. And I was like, well, Cambridge rowing.

Eva-Maria Hempe: The year, my first year I watched the Cambridge boat races and I was like, wow, it must be so nerve wracking and whatever. And then they were like, yeah, you did the summer program. Don’t you want to trial, like just try for the university? And I was like, okay, well, what’s the worst that could happen? I’d taken that lesson of where I hadn’t rowed and regretted it. I’m like, okay, I don’t want to regret. So I just went for it. And then I found myself on the starting line of that boat race, which I just watched a year before. So I went within 18 months from never having rowed in my life to rowing and winning a boat race. And I think the lesson here, as I said, there’s the one about no regrets.

Eva-Maria Hempe: I think the second one about that you’re just capable of a lot more than you give yourself credit for. And I think the third one also just about the power of habits and the power of persistence and the power of community. So there’s nicer things than getting up every single morning at five o’clock, going to the train station, going rowing, barely making it back for nine o’clock to go and to your lab and do your work. And then at five o’clock going back to row. But it’s incredibly disciplining because you only have from nine to five. There is just no, oh, I’ll do this later. You have to be done at five because then you have to leave and go train and you have to be there for training. You can’t skip training.

Eva-Maria Hempe: And so I thought that was actually really useful to fall into this rhythm and go along with it and also shape your environment in a way that helps you do the things you want to do. Because like I said, it’s just not like, I don’t want to get up at five, but I just have to. And then once you’re back from training, you actually feel pretty good. And of course winning the race, nothing feels as good as that. But even if I would have lost the race, I still like, yeah, it was interesting because just before the race, it was about an hour or two before the start. And I remember we were in the boat bay and did like a little circle of the whole crew. And until then I had a bit of nerves, but from that moment on, I was just calm. All the nervousness, all the nerves were just gone.

Eva-Maria Hempe: And I was just like, well, I put everything into this I could, I have no regrets. So whatever happens now on the water, I can look back at this day and I’m proud because I did whatever I could to get to this point. And I think that was interesting because the year before I thought those people must be so nervous when they sit on the start line. But actually when I sat on the start line, I was just calm, I was just ready to do this. And basically put in the work.

Grant Belgard: Why NVIDIA, what sealed the decision for you to join?

Eva-Maria Hempe: It’s because we are a $4 trillion company. No, of course not. Actually, when we joined, I wasn’t. When I joined NVIDIA, it wasn’t a $4 trillion company. No, it’s just, I couldn’t imagine another place right now where you’ll have this impact on the entire ecosystem of healthcare. We work with everybody. We’re the one AI company which works with everybody else. So I get to work with startups. I get to work with established companies. I’m on the forefront of what’s possible. And at the same time on the forefront of what’s possible to do an organization like the thing we thought before. I mean, on the one hand side, we’re looking at models which can design proteins based on 3D structures.

Eva-Maria Hempe: But on the other hand, we’re also looking at rolling out procurement agents because that solves a real problem in the organization today. So it’s just a really exciting place to be at the center of the action around AI and healthcare. And so in general, it just felt like a place where a lot of the things I’ve been doing in the past sort of all came together. Like the multi-stakeholder management of the forum, the strategizing of almost 10 years in consulting, the operationally leading a team and helping people and creating strategies and tactics to make your number, which I did at VMware. And yeah, it just wrapped into sort of this one package of doing something really exciting and really exciting in a field I’m super passionate about.

Grant Belgard: For early career computational biologists who were looking at entering industry, what three skills should they cultivate now?

Eva-Maria Hempe: It’s a bit difficult to say because I’m not a computational biologist, but I think it’s also maybe not so much about the computational and the biologist. I just assume people are well-trained in those fields. I think what’s really important is for them to listen, to sort of to listen where the problems are, what’s being done, where people struggle with. I think the other thing is to really understand value. So I think there’s a lot of interesting work. If you want to do really cool and interesting work, and maybe it’s a bit controversial, but then academia is the place to be. Like if you just are in for the cool, by all means, that’s what academia is supposed to be. If you’re going into industry, then you need to have a nose for value. You need to start to understand like what’s value.

Eva-Maria Hempe: And value can be very different things. Value doesn’t necessarily mean the biggest grossing drug. It can also just be in line with the research portfolio of the organization. It can be in line with individual values of particular managers, but you need to understand value. I think the last thing it’s about teamwork, because so many of these things by now become so difficult that you just can’t solve them alone. You’re dependent on working with others who are bringing complementary skills and complementary experiences. So I would say three things are listening, understanding value, and working well in a team.

Grant Belgard: For life science founders, when is it worth building their own models versus taking existing models or platforms?

Eva-Maria Hempe: So I think you have to be smart. So do you really have an edge? And AI, in my mind, I always think about in three elements. The one is data, compute, and algorithms. So compute, there are some people who have an edge because they can just buy compute for billions of dollars, but that might not be your edge as a founder. So then it probably leaves either algorithms or data. And if you have something there, yeah, you might want to go for it. But very often, actually, you don’t necessarily need to build a model from scratch. You might not even have enough data to build a good model from scratch. And it might be much more worthwhile for what you’re trying to do and you’re coming back to the point of value. What is the value you’re creating?

Eva-Maria Hempe: It might actually be better to stand on the shoulders of giants and just taking a foundation model and retraining it. And in general, I would always advocate for using frameworks out there because they make your work easier. So BioNeMo is not a model per se, but it’s also a framework which helps you do your models better. And I think you shouldn’t write your own data loader and you shouldn’t have tried to configure guardrails from scratch. Like you have, as a founder, you’re massively resource constrained. So try to think about what are the things where you can really differentiate and focus on those and then try to use platforms, existing tools for all the rest.

Eva-Maria Hempe: And I hope that people are taking something from this podcast is we have so much things out there which we’re putting out there, usually often as open source. We have frameworks and libraries and NIMs and all of this is intended to help you and avoid reinventing the wheel. Like if you’re doing medical imaging, you don’t need to write your own segmentation tool. Like this is all out there. Take it and then build a killer application on top of it. But be smart, look at what’s out there and NVIDIA can offer so much and your favorite AI engine, if you ask it, I have this particular problem, what are the latest NVIDIA frameworks? It should give you a whole list of libraries and frameworks you can use, whether it’s for data science or data frames, et cetera. There’s just so much out there.

Eva-Maria Hempe: I think the last thing for life science founders is as well look into Inception. So Inception is NVIDIA’s free virtual accelerator. So it gets you access to NVIDIA experts, which help you even better find the right tools and right frameworks, which make your money last longer. It gets you into a community of like-minded people and there’s also some programs about cloud credits and or discounts for hardware. So join Inception, look at what NVIDIA has and other people have put out there before you build it yourself and just be really smart about what really drives value.

Grant Belgard: What’s your boldest prediction for AI and drug discovery over the next five years?

Eva-Maria Hempe: I don’t know if it’s five years. I would hope it’s five years, but I think at some point we will look back at the way we do drug discovery today and it will seem as archaic and plainly said stupid as the alchemists trying to turn lead into gold. Like today, if you tell kids, oh, back in the middle ages, you had all those alchemists and they were cooking and the idea was lead is this less noble material and you can turn it into more noble material as gold. People are like, why? And I think we look at the same way a lot of things we do today in drug discovery and we’re just like, why did everyone ever think this is going to work?

Eva-Maria Hempe: And there are like on a more practical level, there’s really smart and really interesting things going on about virtual cells and like better predicting like the link between the genome and actually how cells behave. And then also not just cells because we’re not just cells, we’re whole tissues. So I think we’ll see a lot more understanding and understanding biology, at least to some extent. And I think that will get us to this point of alchemy and how could we have been so stupid.

Grant Belgard: What’s a learning resource you would recommend for every trainee?

Eva-Maria Hempe: I think it’s not a learning resource in the conventional way, but I would really encourage to go on build.nvidia.com because it just shows you what’s possible and you have all those different models and you can play with them, you can get an idea what they can do. And then you can also go to the blueprints and basically see how these are put together. So I think that’s a great resource. And then I would maybe pair that with like, I’m a big fan of perplexity, but also any other LLM agent of choice. I think they are great teachers. They can teach you anything. And the other day I used perplexity in voice mode. And so I was like making dinner and just having this really natural conversation. And there is no stupid question. There is no judging.

Eva-Maria Hempe: You can like ask it anything like just, can you please explain to me again how this works? And I sometimes also use it for some of the NVIDIA stuff. I’m like, okay, can we go deeper on RAPIDS? Can you explain the different libraries? Like how does this work? Why does this work? So I think it’s a great tool to learn about AI, but also just anything else you wanna learn. And it can also challenge you. You can actually also ask it to quiz you and to make sure you really understand things and you explain it back to the machine. The machine actually gives you feedback whether you got it right or you need to brush up a bit more.

Grant Belgard: Yeah, I was actually doing the same thing with a bit of yard work yesterday. Also highly recommend that, voice mode is great. Eva-Maria, thank you so much for joining us. It was great.

Eva-Maria Hempe: Thank you, I really enjoyed it.