October 14, 2025 | As drug development pushes into more complex antibody formats, the big question is: how do you really assess developability? In this episode, host Nimish Gera speaks with Shamit Shrivastava, Ph.D., founder and CEO of Apoha, to dive into Liquid Brain, his interdisciplinary innovation that’s changing the way researchers tackle developability of challenging antibody-based drugs. Shamit breaks down what makes Liquid Brain different from traditional developability methods, reveals how it’s unlocking new possibilities for complex format development, and shares a glimpse into Apoha’s future-including an exclusive preview of what’s coming at PEGS Europe this November. 

GUEST BIO

MODERATOR BIO

TRANSCRIPT

Welcome to The Chain, the podcast exploring the lives, careers, research, and discoveries of protein engineers, scientists, and biotech professionals. We look at the impact their work is having on the field and where the industry is headed. Tune in to stay up to date on the newest advancements and to hear the stories that are impacting the world of biologics.

I'm Nimish Gira, your host for today's episode of The Chain, and it is my pleasure to introduce our guest for today, Shamit Shrivastava. Dr. Shamit Shrivastava is a mechanical engineer and biophysicist known for his pioneering work on the thermodynamic theory of nerve pulse propagation. His research focuses on nonlinear acoustic waves and their role in signal transmission within biological systems. As co-founder and CEO of Apoha, a deep tech company, Shamit has transformed his scientific insights into a revolutionary computing platform called the Liquid Brain, applying physical reservoir computing to solve critical challenges like antibody developability. His vision is centered on translating his fundamental discoveries on dynamic properties of soft materials into foundational technologies that tackle seemingly impossible problems in biophysics and material discovery. Welcome to the change, Shamit. Very nice to meet you, uh Nimish. And yeah, it's a pleasure to be here. Can you talk a little bit about what you do currently? And we can go into uh details later, but you can uh maybe introduce yourself as well.

Absolutely. Like I mean, you kind of covered it pretty much. Uh, but yeah, I'm Shamit, and my focus right now, maybe I should talk about that, is on preparing for the explosion of data demand that's, I think, is about to hit biology. And that's really what we are focusing mostly here at uh on Apoha. And uh, I mean, you must be hearing about groups like Cyra, Chai, I mean Charles Dean Group published this really good paper recently. We essentially, you know, it's becoming clear that uh we are now able to push predictive affinity models and generating massive data sets there. And I believe the next bottleneck won't be sequences or affinities, it will be ultra-rich property data produced at the same pace uh that models are learning to consume them. So that's really what is really exciting uh for us, and that's where most of our focus really is right now to get richest data fastest uh on properties of antibodies.

Great. That sounds really exciting, and we're gonna delve deeper into the science there. But I guess before we get into that, our our listeners are always very keen to know about the career trajectory of scientists, and it'll be helpful if you can talk a little bit about how and where did you start your career and what brought you to where you are today.

Oh, yeah, where do I start? So I think like uh the best way to say this would be that my need to actually continue the research uh that I was working on back in academia is actually what led me to start a company eventually. So yeah, I had I had ended up working on something so niche and pro maybe provocative that even academia found it hard to fund, or at least I could not manage that so well. And I believe though that it had sweeping applications, what I was working on. So to truly demonstrate that we need we needed essentially automation and massive data generation, something I think academia doesn't support that well, and I wasn't just going to let the field I was square heading die. So back then the idea was that we were explaining um or exploring whether mechanical vibrations could serve as a new pillar of signaling in biology itself, essentially a new dimension no one was really looking at.

And as a proof point, we really worked on showing that neuronal impulses or nerve impulses, and you kind of touched upon it uh in my introduction, can be explained not only through ion channels, but also through passive propagation of elastic waves in membranes. Now, much like sound waves that we are using right now to talk to each other, and it's traveling so effortlessly in air, can such waves actually play a role in biology at a cellular level? That was like really my work. And what I ended up doing was that uh I demonstrated this in artificial lipid membranes late and that laid the foundation of the technology that we are now actually scaling at Abu Ham. And with that foundational work during my PhD in Boston, I moved to Oxford, I moved, I moved to Oxford and I tried applying these ideas in cancer therapy in a way, using ultrasound and shockwaves to open membranes, actually to deliver antibodies. And that's when I realized how unpredictable antibody behavior really is and how hard it is to push them across membranes without actually killing the cells.

Uh ultimately I couldn't do that. And uh what we ended up doing was we designed nanoparticles to deliver mRNA that expressed K RAS targeting antibodies, our original therapeutic. And uh, and around that time in 2016, I actually then essentially began uh appreciating the huge challenge of antibody developability, I can say, because what I was really appreciating is like how important biophysics of antibodies is to, for example, for delivery of the therapeutic inside cancer cells. And that kind of just exposed me to all sorts of challenges. And I think that was the experience that gave me an aha moment, the same way physics I was using to study membranes. It could actually be applied far more broadly and to characterize any material at the fidelity and efficiency of biology itself. And in other words, we could use the physical neurons that I kind of built on these membranes and sound waves and use them as kind of neural network, but embedded in the real world to predict molecular properties. That's the vision we are pursuing at Apoha today now.

Great, great. Well, thank you so much for sharing all of that. And as I was listening to you talk, I couldn't help think about how interdisciplinary this approach is right now. And you know, you were talking about how in academia maybe there were some challenges in funding research that was maybe more based on automation. You said it was uh provocative. I mean, and and cut to today when you know everyone's talking about automation AI and doing things in a high throughput manner. So it makes sense that you guys are trying to think about applying this to therapeutic protein discovery and characterization. And, you know, that is relevant to the audience for this podcast. So you know, I think it'll be helpful to delve deeper in the development of novel antibody formats and the role you guys are playing. Of course, you described a little bit about the work that you've done previously. But as I was preparing for this podcast, I was looking at this your website and your technology seems to be very interdisciplinary. Can you talk a little bit more about the Liquid Brain technology, as you guys call it, and and really how you identified developability as a key application for this technology? Because it seems like it could be applied in many, many different areas. So maybe let's start there.

Absolutely. It's very exciting. And yeah, I think it's just sometimes there's just the right time for right kind of innovation. And it's it's amazing that you know, all these uh, you know, extraordinary advances that we are seeing in AI, its role in antibody uh discovery and development, and at the same time, high throughput uh screening and need for data generation and this realization of multi-parameter optimization, it could not have been a better timing for us to build a technology like Liquid Brain. So, first of all, just to connect to what I was saying earlier, Liquid Brain is essentially uh the machine that enables us to orchestrate events around this synthetic neuron I was talking about earlier, right? So you can almost think of it in terms of analogy of if if uh if transistor was the uh start of computing, then you actually have to build a chip and then a uh and then a desktop around it to actually make it a complete computer. For us, it's kind of a similar thing. We have the synthetic neurons, which are literally a liquid. These are excitable, that's where the waves and all that uh interactions of your material with our substrate really happen. Um, and around that, this material is staged inside what you can call a processor or a chip, which essentially, you know, kind of help us um orient the system in the right way, maintain it at the right temperature, you know, uh enable us to kind of introduce your sample at a desired pace and desired condition, desired temperature.

So all the like the management or the orientation of the event or the phenomena that you actually want the Liquid Brain to experience. I mean, that's what we call the Liquid Brain, the processor. And then it actually sits inside, you can say, uh like the uh desktop version for us, which is which would be lab robots or HPLC instruments, essentially anything that is designed for automating liquid handling is can be a desktop for us in which Liquid Brain can be embedded. And so essentially that's the kind of stack. And what Liquid Brain is really doing it's essentially it's our way of bringing the principle of AI into the physical world. When you can think of Liquid Brain is essentially like a neural network, but instead of being simulated in a software, it's actually embedded as an excitable substrate in the Petri dish itself, if you may. Um, and what that means is that that substrate naturally processes molecular interactions, the way neurons uh process signals in, I mean, and I say when I say neurons process signals, I'm kind of look thinking about both how they process signals in our brain, but also how artificial neural networks in AI process signals, right? So it's kind of a marrying of two words here that is happening. And uh essentially by processing the signals, if it it is giving us a way to sense, classify, and compute directly from a physical input. So when we thought about where to apply this kind of novel uh technology first, uh, as I was saying earlier, um, I had this aha moment already during my time in academia and I came across the problem of antibody developability. It really stood out as the perfect proving ground. I mean, as we will all know, it's one of the highest stake challenges in drug discovery, where speed, efficiency, and accuracy all matter at once. And at the same time, it's incredibly complex.

You need to read multidimensional molecular fingerprints or properties and make meaningful predictions early when material is really scarce. So essentially, developer gave us a chance to demonstrate both sides of what liquid print can do. It's sensing capabilities, capturing ultra ultra-rich molecular fingerprints, and it's computing capabilities, classifying them into predictive insights. And in many ways, it's one of the hardest benchmarks for material computing. Me coming from material science background can can attest that. And showing we can meet that bar sets the stage for applying the same approach to a whole range of problems beyond antibodies, because uh I think we'll all appreciate uh biophysical properties or um, you know, behavior of materials is kind of a horizontal theme across modalities and not just in pharma, even beyond.

Yeah. Yeah. Great. I mean, again, I mean, I'm I'm amazed by how uh deep physics application is now being applied into the inner body world. And I think, you know, I couldn't help but compare it with other developability assessments that we do currently in industry. I mean, one thing I can see, or maybe I'm wrong here, but maybe you can clarify this, is that it looks like almost the Liquid Brain is like a living and breathing membrane, right? So you're interacting your test sample, which in this case would be an antibody, with something that is very much like uh almost like a tissue, right? So um, can you elaborate how your technology compares to or is differentiated from other developability assessments that scientists can do today in their lab?

That's a great question, Nimish. And I almost want to give like two different kinds of answers, but let me start with the straight answer first. So I think uh, I mean, if you just talk in terms of like straight numbers and value proposition at this point, I mean, liquid paint is essentially, as I said earlier, of AI first measurement platform that captures high-dimensional biophysical fingerprints from a tiny amount of materials. We're talking about 10 microgram at 0.1 per mil is like, you know, it's the minimum that we need to actually uh start doing something with liquid paint, which is incredibly powerful. And then we convert that amount of sample into essentially thousands of descriptors. What that means is like if you have used in silico platforms, you're probably familiar with the concept of descriptors. Liquid Brain kind of does that in a similar way. And then essentially, what liquid pain data or these fingerprints really have are hidden latent properties, which are biophysical properties and they are orthogonal to each other, and we have multiples of them. Um, so two things that are really different are basically first of all, we are orthogonal and very high precision. What that means is that our with this data, we can actually do risk flagging uh in a very high precision and orthogonal way to conventional assay. And and I mean, we published this head-to-head analysis a while back and we can present it at conferences as well, that we can combine liquid print data with standard assays uh and this substantially increases uh recall without sacrificing any precision, meaning teams catch more problematic antibodies earlier at lower cost.

And I think this earlier at lower cost thing, at least at this point, is like the key thing for us because there's really an early stage fit here. Uh, because the assays needs, the assays we run needs so little material and runs so quickly. It's practical, it's basically practically uh uh you can move developability as a concept completely at the hit stage or the early uh in early drug discovery, where traditional methods are you know typically too sample hungry or slow. And then still the information that they generate is very low dimensional for any kind of machine learning or AI to really help help you. So I think that's that's really the key point. Like, can you actually just using 10 microgram, which is the kind of material you have at the earliest stages possible, already generate certain flags uh with very high precision that are very, I mean, high precision in the sense of predicting failure uh with a lot of likelihood. And so that's what we do. And in the end, ultimately it's complementary also because essentially that's literally almost no material. And you can continue to do whatever you are doing later in the pipeline, and you can always then use this data you generated earlier to make better decisions at any stage uh in uh moving forward. So that's the kind of uh I think the key uh aspect of the technology today. Uh does that kind of cover the answer?

Yeah, I I think so. I mean, I again I'm I'm intrigued about what you're describing. So as you describe it, I sort of start thinking about how you know someone in a lab would apply to it and how it is relevant to what's happening in drug development today. So as you probably know, you know, the whole antibody field and therapeutics field is starting to move towards more complex formats. You know, we have more and more bi-specific antibodies being approved now, you know, they are built upon different uh formats and of course single domains, SCFVs, all of these different pieces that people are putting together to generate complex formats. And, you know, the the developability assessments that were sort of put in place almost like over a decade ago, you know, when the uh work from Adimab started coming into play. And of course, the classic paper by Jane at all in 2017 describes in great detail on how to analyze data and look at uh the probability of success with uh approved antibodies or antibodies that have been in the clinic. The the um the type of molecules that are now getting in the clinic are more and more these uh bi-specific molecules or complex antibody drug conjugates. How does your technology assess these formats where I think the traditional solutions right now don't have a set standardized way of doing them and people don't really have the right thresholds? So, does your technology allow people to assess these novel complex formats as well?

Uh yes, I mean that's a really timely question, right? I mean, um I think maybe to answer that if if it's okay, I I take like a step back and kind of maybe it sounds like a detour in the beginning, but uh I would like everyone to kind of uh maybe give me some leeway and explain like the thought that comes to my mind. Sure. So I think um yeah, so I think like uh when it comes to, you know, you kind of touched upon it earlier that Liquid Brain is kind of like a living breathing system and it's it's it's uh it's a membrane and uh uh which is kind of like you know, uh gives a tissue-like environment and and all of those things. I think it's it's uh it's a very um tempting uh uh almost proposition that we are somehow able to mimic biology, and that's why uh we are applicable to antibodies or or all these uh like non-canonical formats that you talked about, Namish. Um, but maybe there's something like which I can go beyond Liquid Brain in general, the state of AI in drug discovery or sorry, antibody discovery that I can talk about. So um so I think like when it comes to um thinking about what is the data that we generate, really there is a precedence for it.

I mean, we talk about Liquid Brain and and waveforms and on all of that, but really there is a just like at the same footing, there are like you know, foundational data sets like structure or sequence. We produce a foundational kind of data set, and it's known very well in material science, they are called state diagrams. And I mean, even in in biophysics, uh we we kind of do them uh and and we take them for granted, but we do these titration experiments or kinetics of binding, where we kind of observe the phenomena unfold over a time. So these are in general, these kinds of measurements where you are doing time domain measurements are known as state diagrams, and that's we believe is like the third pillar of this equation. So you have structure, you have sequence, and you have state, uh, which is essentially the behavior of the material. And now I think when we talk about something as fundamental state diagram, um, is what is it doing? It's basically plotting one function with respect to the other. So, for example, you can plot uh pH as a function of protonation and you get pKa from that girth. Similarly, you can plot you know, kind of surface tension as a function of concentration and you get you know a certain uh like uh adsorption isotherms on the surfaces. So these are all state diagrams and they are universally applicable, right? Like even in fact, they are true not just for antibodies or proteins or or you know, like non-canonical formats, but they are true for you know even lipids or or you know oils or whatever you may think, like liquids and and gas, you know, like state diagrams are true for everything.

So fundamentally, our data is very broadly applicable. But like, of course, the meaning is where, you know, like okay, what do these diagrams then mean specifically for for antibodies or or biaspecifics? That's where, you know, essentially um kind of taking a decision based on data for that the system needs to have, you know, kind of we need our calibration curves, we need our uh small data sets and case studies. And I think the way we kind of try to address that. So, first of all, yes. So this was my detour, but uh but like absolutely yes, we are already working on these formats. So uh I think people who uh can go and check, like, you know, we have kind of published the desirable gen benchmarking, we have done that and we have gone beyond and kind of shown how data looks for 200 antibodies, like the original 137. Plus, yeah, plus 100 more from phase one since then. So that is there. And I think like then if you think about properties like viscosity, et cetera, kind of similar ideal state diagram, they are broadly applicable.

So once you have that comparison to uh monoclonal antibodies or how that looks there, maybe that starts helping you form certain opinion about what does that what does our data mean for these other formats as well. And then of course, we always can, you know, if you if you have controls, and I think that's what probably is a challenge for all the assays right now, uh is they're missing standards. So as standards evolve, essentially you can very easily use them as controls on a system and by like looking at similarities and distances, you can form a very good uh in a very similar format. So you can take the monoclonal antibody case study we did as exact as an example, and you can apply almost the same playbook to any other data set, and you can kind of reach the same kind of conclusions if you have the right controls, and you can basically share that them with us, and then we can kind of plot them on the same uh map and allow you to make those decisions.

Are you enjoying the conversation? We'd love to hear from you. Please subscribe to the podcast and give us a rating. It helps other people find and join the conversation. If you've got speaker or topic ideas, we'd love to hear those too. You can send them in a podcast review.

So I guess if I understand the concept clearly, then it seems like you're looking at overall properties of a molecule. So in some way, it's almost independent of the molecule that you're using to interact with the Liquid Brain. It could basically be any form of the antibody, and it gives you the state diagram that describes the overall properties of the molecule that you're testing. Now you could look at these different state diagrams as you described them to understand how the overall behavior of that molecule is sort of um changing, or how the molecules are different from each other when you sort of test a set together. Okay. Okay. Yeah, it's very interesting. I think it's a very maybe different way of thinking about overall properties because we always think of when we're running assays, we run these assays one by one for each property. But in the end, you know, when you're making a molecule, all those properties sort of come together. You know, when you're, you know, formulating it or you know, generating it at a high concentration, those properties all matter and they sort of work together. They don't work in isolation. Exactly.

Yeah, this is a very classic example of, you know, like uh I think uh Jan did a great great job in kind of taking all the assays and kind of building the picture of deliberately kind of bottom up, like, okay, these are the legacy assays. Let's see how what are their outputs and can we kind of create a uh, you know, like a high-level picture combining them. But like you can almost think, okay, now once we know that biophysical behavior is important, where properties are kind of uh kind of you know point um uh point behavior almost. Like state diagram is big thing, like a uh like a bigger representation, and properties are almost like features of that state diagram. I give the example of, you know, from the hydration of pH, PK is just like one feature where the inflection happens, right? So so exactly like the properties are almost like the uh tip of the iceberg if the iceberg is the behavior state diagram. So uh so yeah, essentially, like I mean, um you can almost think like now that everyone is sold that biophysical behavior of state, like almost I would say state diagrams are important. Let's find the best way to generate richest state diagrams in the most efficient way.

And now we can work kind of top down. Okay, let's find out uh kind of in a developability first manner what features of these state diagrams are actually playing a role. And yes, in many cases they will align with the properties we all already know. But we believe in many cases we might find new uh liabilities that we haven't seen before. And we kind of see that in our data sets that we have generated. We find uh we were able to flag risky antibodies. I mean, it's it's uh it's a discussion on its own, but uh in like the discussion on its own is basically what is a risky antibody, but if we keep that aside, we have found uh liabilities in molecules which are missed by traditional essays. So this is a very interesting uh kind of aspect of the work that we are doing that's emerging. And we are collaborating with a lot of experts in the field to kind of get deeper into this.

Yeah, and it sounds like probably what you said earlier, you guys are constantly generating more and more data and then also using AI and machine learning to sort of learn from that data. And as you sort of collect more data, you get better and better and at predicting things.

Exactly.

Great, good. Yes, I think now the big question is for people that are listening and might want to leverage your technology or or work with you guys, how do you typically engage with people? Do you do collaborations with academia or industry or both? Do you do like pilot programs? How should people you know reach out if they want to work and leverage the Liquid Brain technology?

Awesome. I think kind of all of the four. Yeah. Uh so uh basically, I think given the value that we generate is so foundational. If you think of the whole um thing we are trying to do with antibody design, if you think of that as a stack, then we kind of sit at the lowest layer in the foundational layer, right? So we generate data. And I think anyone who is interested in absorbing our data can be of a partner, collaborator, customer, you know, and we work with uh academics, we work with startups, uh, we work with companies who are building models. Um, and and of course, we work directly with our end customer who are people who are designing antibodies themselves. Yeah. And uh all of this is to really build, like the vision is to have this collaborative ecosystem where we can kind of all uh share our values and bring together uh the right kind of tools and the most optimal workflows together. So this can be really accelerated and maybe even democratized. So the way to work with us, first of all, you can work with us as a remote service. Like if you have samples, very easy to just send those to us. You can go uh go to our website and request it.

And essentially uh you can just send us like 100 samples at 0.1 per mil in PBS or HDIN, and then we will receive it. Um, uh we'll run them, and then essentially all the results and not just like a table or report, you get like the entire suite of information within Silico tools that can help you like really even do early stages of screening and and triaging all on a all on a uh software platform that you can actually open in your web browser. So that you can bring all of that together in in our uh software platform and then kind of directly monitor how the results are coming through as they happen and and kind of not just get just raw data from us, but actual insights. We'll uh we will provide uh the recommended thresholds, but you can move them around and kind of uh upload your own data to kind of triage everything together, not just with Liquid Brain data, but any data you have or any hypothesis you have, or if you you know even can run uh generate new descriptors from uh well-established uh and popular in silico uh tools, end-to-end encrypted or secure. So that's like you know, if you are essentially yourself someone who are who is designing antibodies, we essentially provide like really a very solid um proposition around just the developability.

Then there is, you know, uh, of course, academic and ML collaborations, we have already public data sets that we make available to you to test, see if your if your models can improve. We also have run small grants where essentially, if you are an academic and you have you're working on some kind of algorithm to improve antibody properties, we provide our data to you, you can make predictions, and if you have specific prediction that you would you would like to test again, we we take that cost and we help you, you know, kind of generate more molecules and kind of validate the data further and your models further. So there are there are many different ways actually in which we work. And we are, I mean, one thing in the early stages of startup is like that, you know, anyone would uh really value most is like we are really keen and motivated to work with people and kind of you know uh be available and helpful. And I you almost get us as an extended RD team uh or AI team, essentially. We have in-house expertise, like cutting-edge expertise in AI, machine learning, material science, biophysics, computational biology. So yeah, I mean, it's not just the software and and uh and data at this point, you get people in a way as well.

Okay, okay. So it sounds like if someone is interested, they should reach out and you guys will find a way to work together.

Absolutely. Very easy to just contact us on Apoha.com.

Great, great. So I guess for for the Liquid Brain technology or overall for Apoha's future in in, say, the antibody space, what should we be watching out over the next couple of years?

Yeah, I think there are mainly three concrete arcs on which we are working. First of all, right now we give we give flags uh from our data, and I think it's very obvious that we move to what is causing those flags. So, for example, uh moving up beyond our current single risk flag to increasingly explainable biophysical factors, linking our latent properties to you know, like interpretable attributes like hydrophobic liability in this part of the sequence, or aggregation-prone states, or you know, uh doing all of this while preserving this early stage precision that we have. The second arc is really kind of what I touched upon a little bit, maybe in meandering words, but like very much interested in ecosystem and its validation.

So, for example, we want to, or we will be launching the launching a modular developer ecosystem, uh, basically kind of doing validation as a service. We are tying up with CROs so teams can actually iterate sequence prediction and then Liquid Brain validation in tight loops and and you know, continuously improve models and build internal assets almost. And finally, of course, deployment and workflows. So, like this is very transitionary for us that the samples are being sent to us and we are sending data. Uh, I mean, the most exciting thing probably that's coming on soon is like first on frame, on prem deployment with one button workflows essentially. And you know, you thus you can expand assays that interpret, uh, interrogate manufactured irrelevant behavior. Stress robustness, viscosity, link signatures, formulation sensitivity. So a lot of work happening. And I think like one thing I should mention, this 10 microgram, as I said in the beginning, is the minimum requirement to do something amazing with Liquid Brain.

But essentially, in the end, Liquid Brain is not a single, what do you what do they say? Like single uh single shot pony or pony or whatever there's a phrase. Uh uh, but basically you can program it to do all kinds of, you know, uh setting all kinds of interactions on all kinds of stress profiles. You can change pH and to pH-dependent profiles. Essentially, you know, 10 microgram is the minimum, but you can add 10 more and 10 more and start generating new and new kinds of behavioral data sets. So what I'm getting at is like early stage is like this unique capabilities, uh, capability that we have, but what we also will be expanding towards is like later stage uh or like more material um extensive uh measurements that are very hard today. Like, for example, predicting viscosity at high concentrations. And I'm there maybe people are still okay to spend 100 microgram for a lot more data than 10 microgram for part of that data, right? So yeah, those kind of things.

Great, great. I also noticed that Apoho will be exhibiting at Plex Europe in a couple of weeks. And I was wondering if you can tell us anything about any new applications or any new solutions that would be shared there.

Absolutely. I mean, I think the kind of things I touched upon, uh, we will provide you solid evidence of making progress along those lines. So you'll see expanded benchmarks again uh on some of the things I talked about, especially some data set on viscosity that we have done. And then uh also, you know, like we are working a lot in collaboration with some uh key opinion leaders in academia and and industry, where uh essentially sharing a lot more case studies um on what exactly are the biggest problems in the space and how you would solve them on Liquid Brain. So a lot a lot to look forward in those directions and including something we touched upon in the very beginning. Also, how does like you know, you know, like these non-canonical formats look like on Liquid Brain, and how do you use uh liquid train to uh move the needle in those uh in those modalities? So yeah, and then finally, uh like a preview of our kind of the insights or software platform where you can run these type loops of in silico prediction to to validation and to design uh in a closed loop. So, yeah, all of that and more uh in these new conferences.

Great, great. It's been great chatting with you, Shamit, today. And thank you so much for taking the time to speak with us and sharing our experience for our listeners.

Absolutely. It was a pleasure, and I'm always very excited to talk about science uh and and you know, like uh and anyone who's kind of uh resonated with any of the stuff I said today, kind of trying to think about these problems from maybe, you know, from an outsider perspective, which I'm obviously here, but although I've been working on these problems for six, seven years now, uh so yeah, very, very interested in networking with any such voices that uh that are hearing this podcast. Yeah.

Great. And of course, thank you to our listeners now for tuning in. And please join us again as we release new episodes every month. Enjoy listening to this and some of our previous episodes on the chain website or wherever you get your podcast. Thank you, Shamit.

Thank you, Nimish.