Latent Space: The AI Engineer Podcast - 🔬Causal Models Need Causal Data - Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist)

Episode Date: July 21, 2026

Bet on informationIf test loss flatlines after 1.5B parameters while training loss continues to drop as you scale, that tells you that your model is limited by the amount of information in your data.Training on a single, smallish data set exposed an information gap: the 3.1B model falls off the scaling trend. Neither parameters nor compute will improve performance past this wall. For predicting changes to gene expression, you need more information rich data.This is what Chu and Bo’s teams have done, and here is what ~30x the information buys you:Now we can scale with parameters and training compute! We don’t know how much this effort costed, but we can guess that data collection experiments and infrastructure was a few tens of millions, and compute + headcount + research was a few million. The budget looks like a RL rollout budget, rather than a data rich pre-training one.We were lucky enough to have the two central figures in this story on our podcast. Taking the lead from Ci Chu and Bo Wang, Xaira Therapeutics is betting that information rich data is the key to AI-driven drug development. Chu was recently promoted to Chief Discovery Officer and Bo to Chief AI Scientist, underscoring just how strategic Xaira considers this bet.Reverse engineering the human cellIf you had to figure out how a human cell works, what would you do? A good place to start might be by documenting what genes are expressed (e.g. what RNA is floating around) in different kinds of cells, in different circumstances.That is CELLxGENE, a database of 168M cells built by Chan Zuckerberg Institute that maps each cell to a count of how many times 20K-30K genes were detected in that cell, plus detailed metadata about every cell. A ~4 trillion-entry matrix.If the Protein Data Bank (PDB) unlocked structural biology models (Boltz Episode, ESM/BioHub Episode), CELLxGENE has done the same thing for Virtual Cell models. Like PDB, CELLxGENE has inspired a zoo of AI models of RNA expression; so much so that RNA expression models have become synonymous with Virtual Cell models. Bo Wang built one of the most influential, scGPT, that became the starting point for Xaira’s new model.RNA expression ≠ Virtual CellModels trained on CELLxGENE describe the relationship between cell types and cell states, but they are not good at predicting what will happen if we make changes to RNA expression. Changes in gene expression are highly correlated, and its is difficult (impossible) to figure out what causes what in most cases.If you could “turn the dial down” on one gene at a time, however, then you would be able to observe what is upstream and downstream of a given gene. You could tell if A → B & C or B → A & C or B → A, C → B → … If you did this for all of the genes, then maybe you could train a model that could predict what would happen to a cell if you change a gene (e.g. with a drug or a gene edit). Or maybe you could figure out the least invasive way to change a particular gene’s expression.X-Atlas → X-CellThis is exactly what Chu and Bo’s teams have done. The data set is called X-Atlas and the model is called X-Cell.In this episode, we discuss:* Why the team abandoned autoregression for diffusion* The CRISPR-based experiments that run millions of tests in parallel, and generate the raw data for X-Atlas and X-cell* Generalization to real lab experiments in real human cells* Beating the linear baseline that has outperformed previous models* Justifying a kitchen-sink of priors, and how that stacks up vs. data and architectureBo also shared with us some of the (major) advantages he has as an academic vs. industry leader, and how his labs keep up with the breakneck pace of AI innovation.Check out the full episode on YouTube, or your favorite podcasting platform! This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.latent.space/subscribe

And what really blew my mind away is when I saw the model make prediction, just print out the heat map of the genus conversion changes. Look at the actual raw data and line up the linear baseline prediction, the ground truth, and the X-L prediction altogether. It's visually very clear to see that X-L prediction is much more similar to ground truth than the linear baseline. This is a wow moment I was talking about in the beginning. This is the first time that someone can put together not just one per terpsy, but seven genome wire per tripsy campaigns together.

Something that jumped out to us, biologists, right away, is that some of the perturbations are contacts universal. Hi, I'm RJ Hanaki, CTO of Meriomics. This is Brandon Anderson, who builds RNA therapeutics at Atomic A.I. And this is the latent space AI for science podcast. One of the themes that has run through the podcast is how the lab, And experimentation and the real world have probably the biggest impact and have the most relevance to whether something is AI for science or something like B2B SAS. We're really happy to have in the studio with us today, Bo Wang and C-Chu from Zara Therapeutics.

At Zara, they're building with a bunch of other people, a AI drug discovery platform. They're using high-throughput experimentation system to collect very large data sets and then training AI models that can predict the way that your cells in your body will respond to drugs and therapeutics. Really happy to have you, a big fan of your work. Why don't you two introduce yourselves to the listeners? Hello, everyone. My name is Bowen. I'm SVP and head of biomedical AI at Xero Therapeutic. joined there about eight months ago, and before that, I was associate professor at the University of Toronto in Canada. And I'm Tsichu.

My first name is incredibly difficult to pronounce unless you speak Mandarin, so I go by Chu, as in Chubacher or Pikachu. Favorite fictional character. I'm the SVP of AI-enabled Discovery at Zara. I joined about more than two years ago when I was still in stealth mode, and here I lead the high-throughput biology group. generating the kind of data that will feed our AI models and also think about their applications. Before this, I spent about a decade at the intersection of AI and big data and biology. So previously I worked at InCitral, leading the InVitro Discovery platform there. And before that, I was at Verily, which spun out of Google X.

Okay, so you are at Zara, the company, which is on the Pareto Frontier of Confusing Names and Megar rounds. So Zara is, I think, kind of came out of stealth like a few years ago and just really big org kind of out of nothing. So I'm curious if you can explain a little bit about what is Zara's mission? What is their thesis statement? Like what is, you know, special about Zara and, you know, kind of where you're going in the future? Yeah, Zara is a AI-enabled drug discovery company. And at the core of her mission, we're using AI platforms to generate better therapeutics to advance patient care. And so we will be making drugs using different AI capabilities. There are three main AI platforms that we're building here.

The first one is protein design, work that spun out of our co-founder, Dr. David Baker's group from UW. A lot of the current generation of protein designers are here in the company. So there the thinking is to use advanced AI technology to develop molecules against previously undruggable targets. The second AI platform, I guess we'll spend a lot of time talking about today, is the one that Billena I have been working on for quite some time and just release the preprint on. That's the virtual cell or foundation model of biology work. There the hope is to build an AI model to predict biology, exactly like you said, and predict what genes and drug molecules will affect cell biology.

And the third piece, which we're beginning to build now, is patient representation models. And the goal there is to have AI models that can understand which patients will respond to which therapeutics. So hopefully together, these platform technology will help us make better drugs faster and with a higher success rate than previous technologies to transform what used to be artisanal tried in the era in the past into more and more into an engineering discipline. I think what sets there are different is not just the one billion

talking around, but also I think there is one of the very few AI native companies for drug discovery that works from end to end of all sections of drug discovery from as early as you know target ID and the end protein design small molecules and two phase one to three clinical trials we aim to use AI to accelerate every part of the drug discovery. So that not only we increase the success rate of developing drugs, but also greatly reduce the cycle time so that we can have new drugs instead of every 10, 20 years. So hopefully we can have the cycle times that we have more useful drugs for patients.

That's really interesting. I know there's a lot of interest right now in that third thing, maybe called translation, from the lab to the clinic. Where are the bottlenecks? So you have these three models. What are the bottlenecks that you're addressing? And sort of like, how are you doing that? Why are you doing it that way?

There is an AI native company. Almost every part of the sections of drug discovery, we're trying to use AI to revolutionize how we develop drugs. So the early part, we build causal foundation models, or sometimes we call it virtual cell. proteins, we have, you know, state of arts, protein engineering models, and we have also patient representation learning models. And I think what Zara is trying to do is not only we develop AI models, but also we create the right data sets to empower these models. And I think what

really makes me excited to work at Zara is we always aim to connect three AI models together, instead of letting them work individually by their own. So when we design virtual cell models, we look for connections to that, can we find targets that is easier to apply the protein engineering models? And then even when we design the cellular causal models, can we connect to patient representations? What are the right patient data to connect to the cellular models

so that we have something to show clinical utilities? So I think what really make me excited is before I join there, I'm kind of a professor in computational biology department or computer science department where we're mostly working on computers, we look at the data, look at arrays, etc. Once coming to Zara, what really excites me is that I get to talk to people like too, lots of drug hunters, you know, extremely experienced drug hunters to really understand their pinpoint. So when we do that, we do that, design AI models, we think about questions that really excites biologists. So later, maybe we can talk about how one of the rewarding signals I receive after we develop

Excel is that like the wow moments from biologists that this is the first time biologist actually find the model can predict exactly how these unseen cell lines kind of respond to different perturbations. So that's kind of the part really excites me is the integration. of kind of dry lab or AI models to wet lab or the biology or even eventually to the clinical side. With this clinical model, I know you guys are aiming to, you know, take a drug all the way to FDA approval and beyond. Where do we stand now? I don't know if you're able to talk about this,

but like are you able to collect data from clinical trials and tie that back yet? As both said, I think if you think about drug discovery process, it's easy. right? You just need to find a right target, make the right molecule, and find the right patients to give them to. Of course, each of those steps are incredibly difficult to get right. And so far, like I said, just now, it realized a lot of untrodden error in guesswork. And the main issue, I think, is that we don't have to write biological data, really the power the training of a predictive model. And in protein design space, I think that's where we have seen the most rapid progress so far.

That's partially because we have a lot of data, high-quality data, over 70 years, curated by the entire community, people deposit protein structures into a database called PDB. We also have a lot of sequence data collected over the years from different genomes that can help inform the model as well. And it's these high-quality data that are collected and accumulated that ushered in this revolution in a prismine design and alpha fold and other folding models. In the other domains, such as clinical model prediction, such as virtual cell, we are nowhere near the same kind of massive data that are high quality.

And I think it's mainly a data limitation issue. So to your question, that's where we're invested in generating these data, particularly causal data in cell biology in a lab. And that's, I think, what made it possible to innovate on the algorithm side. as well to usher in virtual cell models. On the patient's side, it's a very interesting question. Perhaps that's one of the hardest data to get because getting access to high-quality patient samples

is difficult in itself. Getting it matched to the right clinical annotation so that you can actually learn the difference, the bridge between molecular data and clinical response, that's even harder. And you might be able to do that across different disease severities, but it will be harder to collect the right data to predict which drug treatment will or will not respond in a particular patient or not. And so that takes a lot of thought

and a lot of careful curation to generate data out of. So we're beginning to go into the area, but hopefully we'll be able to share more soon. Awesome. Maybe we should switch gear. Now, you just released X-cell. Why don't you guys describe? I'll butcher it. X-C-cell is Zara's first virtual cell models. It is an AI model that can predict the response to genetic perturbations. Certainly, we can extend it to other type of interventions such as drug perturbations, chemical perturbations, etc. So can you just describe for the non-biologists that are listening?

What is a perturbation? What do you mean by that? In our cells, when we'll talk about genetic perturbations, our cells, human cells typically have 20,000 genes. Not all cells express every gene equally. That's why your eye cell, your skin cell, your heart cell, even though they share the same genome, they function very differently.

A lot of that is determined by selective gene expression that determine the type and the state of the cell. So what we do is to build a model that you can in silico ablate certain genes from the cell, That is an encyclical perturbation. That's to say, if I reduce the expression of this gene in a cell, what is the implication for the rest of the cells? What's the biological consequence?

You basically turn the knob down on one gene. That's right. And then what happens to all the other genes in that cell? Correct. And the hope is, of course, to predict the effect on all the other genes, but maybe even more things than gene expression, such as the function of the cell. Okay.

And that's important that that's therapeutically relevant because a lot of drugs are inhibitors, and they function through exactly that, turning down the activity of a protein or gene. And so we can start with gene, perturbation prediction. The hope is that we can also go to pathway inhibition prediction, so on and so forth. So a pathway is just a set of genes that all kind of talk to each other by this gene expresses a protein. That protein has some impact on another gene and so forth and so on. there's this long chain reaction of genes and proteins.

And then so that that's called a pathway. And so if you interrupt that or somehow change it, then that has an impact on the larger phenotype of the cell, what the cell looks like does, et cetera. That's exactly right. Yeah. Yeah. So you have what you call a virtual cell or you're creating a virtual cell. And virtual cells are very popular. A lot of people are interested in this concept.

But I think your approach is somewhat unique. or separate from other people are doing? Can you explain what do broadly people mean when they say virtual cells? What are some of the distinct other strategies? And then like what is your specific strategy that you're going for? Certainly virtual cell is a very high level term to describe an AI model that is able to predict or describe what cell looks like or predict the cell expressions or cell functions after certain interventions. It's a very high level concept. It was, first of all, it was not a novel idea.

We had virtual sale project almost 20 years ago. But back then, sometimes we call virtual sale 1.0, is that people are trying to derive differential equations to try to use mathematics to describe what's a response for certain pathway intervention, as you just mentioned. And by feeding these equations to different observations. And largely speaking, that was a failed attempt in the sense that the biology is just way too complicated

to write in a few predefined set of differential equations. Moving forward, with the rise of language models, I think the idea of using AI models to mimic how cell responds to different interventions by data-dreaming approach, I started to get popular. And I think three years ago, almost just four months

after Chattvi was released, our lab at University of Toronto published one of the early foundation model of single-celled genomics, called it SDGBT. It can kind of interpret it as a GBT-like model for single cells.

And it quickly become very popular in the sense that for the first time we have a foundation model that is able to tackle different downstream tasks using the same model, such as we can use the same model to integrate different batches of single-cell RNIC. We can use the same model to predict multi-omic integrations. Let's define those things. So batches integrate different batches of RNA-seek. So you have different equipment. You're all collecting. Maybe you're collecting the same.

Different labs. Yeah, different labs, different time of day, different phase of the moon, whatever. And those actually have a big impact on the data that you call. And so there's a big problem of how do I even compare this data set to that data set when there's all this other differences that have nothing to do with the gene expression and just how I measured it. We call that batch effect. We certainly want to remove the batch effect while preserving the cell types, which are more important biology we want to reserve.

So this is sort of like analogous to the tank problem in image classifiers, for example. It's sort of the models pick up on these crazy spurious features, which have nothing to do with what you actually care about, the underlying biology. Certainly, the core idea of integrating different batches is to kind of keep the biological signals while removing the batch effect. And before these foundation models, what happens in single-cell domain is that for every task, biologists have to choose the so-called specialist state-of-art approaches. And with foundation models such as SGVD or gene formers, what we hope to bring is that one model that solve all the tasks in single cells. And with the popularity of foundation model, lots of researchers come together under CZI, Chen Zagreb Institute, and we published a prospective paper as journal Cell to coin, for the first time, coined the term virtual cell, almost virtual cell 2.0, in the sense, let's use, data-driven approaches. If we cannot describe, let's learn it. So that's the idea of virtual

cell so that genus being can we build a language model or language type of model to predict what the cell types looks like, how the cell respond to different interventions, and eventually we can replace all the cellular experiments by simply running simulations on computer without even running the actual experiments. Maybe for a bit more context, you can think about this is... So a virtual cell is just a general concept, but you think cells have 20,000 genes in them. And in most human cells, I think,

what roughly, you know, 4 to 5,000 are usually active at any given time, or expressed at reasonable levels. So you look at a normal cell, you might have 4 to 5,000 genes doing things. And so your question is, in many cases, the way you... medicine works is you, you know, you target a protein or you target some sort of, you know, something which makes proteins more common or less common, or they stop the protein from doing

something. And your goal is given this, you know, some number of genes which are in a cell, every cell has a different composition of genes. What is going to change? You know, will some pathway die off? Will some pathway grow? And how does this, you know, from this, you know, from this, you could predict how medicine is going to work by just understanding how changing one specific gene or some cluster of genes could change everything. Is that a correct understanding? Yeah, that's a correct high-level understanding about virtual cell. What's happening for this field is that we are lacking a concrete definition of virtual cells. And people almost equate foundation model with virtual cell.

But in my view, virtual cell is probably a much. broader concept than just foundation models. Foundation models mostly provide a reliable, semantic, meaningful representations of cells. But I think virtual cell is more dynamic in the sense that can we build AI models, even predict the development of the cell states across different times, or can we even describe the spatial changes at the cells, different cellular resolutions? In my understanding, is that we are really at the early stage to develop such comprehensive virtual cell models. And the foundation model is really just the starting point.

AI models always begin with the data. You are building a high-throughput experiment or have built and are continuing to develop a high-throughput experimentation system. Can you, that sounds really cool and really complicated? Can you tell us what that entails? What are you doing? What are the experiments? that you're running, how does that inform the building of an AI model and why do this rather than pick up the cell X gene database, which is a collection of gene expression data that

has been aggregated over the public data sets? Yeah, great question. I want to pick up where Bill left off. I think both says something pretty profound, going from a representation model, the foundation model of biology, to a virtual cell. And the key difference there is perturbation, prediction or dynamic processes in biology, that's a causal concept. For that, I think we need causal data.

And if you look at cell by gene, that's a fantastic data set that curated in the beginning more than 33 million cells now a lot more than that. And at the time when SGBT was trained on that data set coming out of both lab in Toronto, So that was mostly an observational profiling data set. It's a descriptive data, not causal, and mostly profiling healthy human donors. And so the model that was trained on this data set is very, very good at doing descriptive tasks, such as harmonizing across batch effects, removing effects from different labs, different technologies.

But I think both us and many others in the field have found that these models that are trained on descriptive data do not yet outperform linear models on causal tasks, perturbational tasks, what we call counterfactual tasks. If I did this to the cell, then what would happen? That makes intuitive sense to a biologist because the correlation data in the descriptive data set can be fit with many, many possible causal structures. In a very simplistic case, let's say you observe gene A, B, C, all go up and down together in your descriptive data. In your descriptive data set. You can infer that A regulate B and C. That's why when A goes up, B and C also go up. You might also say that B regulate A and C, and that will be perfectly reasonable as well.

You could also say that A, regulate B, and C is completely regulated by something different. You see the problem there. And there's N number of way to fit a causal regulatory network into the scripted data. Fundamentally, we believe our original data are underpowered to learn causality, truly. And this is why we realized pretty early on that we need to really start training causal data set to train a causal model. So what are the ways to do that? I think the field has come of age to do these at-scale technique that we call high-thruple biology. And there are many ways to generate these causal data at scale. The technique that we have focused on is something called perturbs-seek. So for the listeners who are not familiar with that,

technology. It combines high-throughput pulled CRISPR perturbation together with single-cell RNA-Sic technology to build 2D datasets. Let me break that down. Yeah. Please. So we just talk about in a cell, there are at least 20,000 analysts to measure. These are the genes. These are both the features to measure. These are also the lever to perturb the cells with. So for clarity, let's call them perturbations and genealicate. expressions for perturbation on one axis and the features that you measure that describe the cell on the other axes. Perturpsi is a technique that leverages the latest breakthrough in lab biology, CRISPR-C-9.

These are bacterially derived enzymes that allows you to disrupt gene expression in mammalian cells, in human cells, for example. And we can do so in one. at a time fashion, so I can take out one gene at a time. Of course, that would be incredibly difficult to scale if I want to do all 20,000 gene expression in that kind of in one single experiment. I probably need a huge factory, a lot of robots to do that. Or you can do them in a pool fashion, and I love pulled experiments.

These are hyper-scalable. So we have lab tricks that allows to disrupt one gene per cell, but do all 20,000 genes across many, many cells in one single pulled experiment. Perfectly scrambled, so there's no batch effects, there's no play-to-play differences. So that's all of the perturbation throughput. Basically use some sort of commutorial trick to first perturb all the different genes

and different combinations, and then you can read them out and do some math on it, and you basically pull out a whole bunch of different experiments in one experiment. Correct. It requires barcoding technology, and that barcode is actually achieved by directly reading out what kind of CRISPR guide RNA is present in which cell. So for CRISPR-Kast9, this bacterially derived machinery to work in many many cells, you just have to deliver two things to each cell.

You have to deliver the protein, the Kastain protein, that does the job. And you have to deliver an address barcode encoded by a short piece of RNA called a guide RNA. And the guide RNA tells the protein where to go in the cell, purely via Watsonquake-based pairing ATCG. So it matches a part of the gene. It's sufficiently long to say this will match the correct gene, and then that guides it to connect to the right and reduce the expression of that particular gene in the cell.

Correct. We designed this guide to go to the promoter part of the gene. That's the beginning stretch of every gene before the transcription starts. And if we bring the Cas9 protein to there, arm with the right effector, the silencer, that promoter would get shut off and that gene will never be transcribed out of again. So effectively we'll tune down the expression level of that gene. And so all you have to know is figure out which guide RNA is in which cell. And that can be done using genomic redouts.

That's the barcode. And you can then infer which gene is being silenced in which cell. So that's the way you scale a throughput on the perturbation side. On the redout side, it's a 2D dataset, right? So we just talk about one of the dimensions. On the redoubt side, we leverage single cell RNA-SIC technologies. So these are also recent technologies in the last decade that have been scaled that can let you read-out the expression level of all 20,000 genes simultaneously from each cell.

So armed with both high-thru-put, CRISPR perturbation and high-throughput single-R-NIC technologies, all of a sudden we can generate these 2D datasets where we systematically perturb or knockout, knock down, every single gene in the human genome in the cell type, and we read out this impact on every other genes in the same cells. So we generate these 2D rich data set not that different than the type of PDB data that trained alpha-phone models, right? If you think about that, that's hundreds of thousands of protein entries. If those are the rows, columns are the XYZ coordinate of every single amino acid. That's also a 2D dataset. And I think it's these type of rich 2D data. And I think it's this type of rich 2D datasets that power the training of foundation models of biology.

I find it really fun how you have turned a fairly straightforward assay in using, this is NGS sequencing, right, next generation sequencing, very high throughput. You've used this to scale a simple perturbation response, which is independently maybe not all that interesting to this massive scale of over basically an arbitrary number of cells. I think you did 24. million or something. So it's actually a lot more than that. So 25 million is what came out of the most stringent quality filtering. It's actually as much of a scientific challenge to figure out how to do CRISPR and Singles R&AC as it is an engineering challenge. In the first part of the experiment, oftentimes we have to harvest tens, if not hundreds of millions of cells and they go through various

quality funnels to arrive, to give Bowling Team the highest quality data at the end. That's incredibly difficult to do because as you can imagine, all of these techniques have been published by academia before and they work very well in small-scale experiments. But when you think about scaling them to a genome-wide perturbation, we're talking about handling hundreds of millions of cells. Techniques that are publishing academia used to be all about handling fresh cells. Cells are still alive. And that may be okay if your entire experiment takes only an hour or two.

It's not quite easy to handle cells across a 14-hour day that's hundreds of millions of cells. And so by the end of the day, he's the Jokwma team. You can easily detect stress signals from the cells and from your scientists in the lab. And quickly, we realized that's not the way to do these data generation. Machine learning is very quality dependent. And we want to give the highest quality data to our AI teams. So we're putting a lot of engineering thought and industrialize the whole.

workflow, step by step, introduce chemical fixations so that we lock the state of the cells in at the beginning of this experiment, but figure out ways that it doesn't disrupt all of the biology, molecular biology steps afterwards. It doesn't impact data quality so that we can do all of these data generation in a time shifted operational manner that's very not prone to batch effects. One thing that you didn't mention is that you're using some sort of stem cells. And so And obviously, like, you don't have, you know, like brain cells or or blood cells. Or if you did, then you would have a big commentatorial effect on that. So how are you, why are you convinced that working on stem cells, which are, you know, my understanding is that there are actually blood cells that have been sort of the stem cell behavior has been unlocked on them.

And that causes some sort of stress on the cell as well. So you have these like sort of not quite blood cells that are stressed. And then why are we convinced that that is a good proxy for a brain cell or whatever you're studying? Yeah, not quite. So we didn't actually start with stem cells. That was more of a later development. When we started data generation, so we put out the method that I talk about, as well as the first two data sets, which is the world's largest perplexic data release at the time.

last June in the preprint, we call a dataset X-Ls-Oryan. That was actually generated from two cell lines. A lot of this field's early work started with cell lines. These are cells. One of them is a cancer cell line. The other is just a cell line. These are immortalized cells. Some of them are derived from cancers, hence cancer cell lines.

Others are just derived from primary cells but have been immortalized, many times grown for many, many years in various labs. people start with these cell lines in the beginning, as you can imagine, because those are easy to do. It's easy to scale, easy to grow a lot of cells out of. Turns out the ability to grow a means of cells is actually critical for doing these large experiments. So we started there first, and they actually still capture the characteristics of the cell types that are derived from, colorectal cancer, as well as hematopoetic cells. But later on, in the most recent preprint,

We actually expanded to many more cell types. Now, some of these are still cell lines. Our T cells, we chose to use cell lines. But some of these have now gone into primary cells. So we did one experiment in IPSC. These are induced plopoten stem cells. And another experiment in, and we think this is the most ambitious and coolest screen that we've done to date. This is a pan-differentiation multi-cell type stem cell project.

So effectively, we differentiate IPSC into. 10 different cell types in one single experiment without restriction. And we did a genome scale perturbation across them. So you can imagine instead of just generating 10,000 different biological experiments, we did 10,000 by 10 cell types, it was almost a library-on-library experiment. Why are we doing this? We think that in the beginning phase of data collection, as both said,

I think we're just in the early days of virtual cell building, context and diversity and richness of the data matters. It's not just a total number of cells or total number of sequencing reads. It's about bits per dollar and information content. So we want to scale not only in the genetic prohibition landscape, but we also want to scale across biological contexts so that we can give our AIR teams the best rich data set to build a generalizable model on.

Is there any thinking about, so you say context, But obviously these cells and these experiments have been sort of D, I don't forget the term, but they've been separated from their cohorts, right? Is there thinking about using spatial transcriptomics or other, you know, sort of technologies, imaging-based technologies to build models with perturbations, but in the context of the cells that it lives near? Great question. So we're thinking about that in a couple of ways. Number one, that's actually exactly why we want to build them. cell model in the first place. You might think that, well, you can already do exhaustive screening in these cell lines. Why do you still need a model? You can just do the experiment and generate

that data. Certainly, if your query is just about cell biology in cell lines, you're right. We don't need a model, right? At least if for genetic screening, we can just do the experiment. But you're also correct that oftentimes good targets, biological insights are not about cell lines. These are about primary cells, about cells in their native physiological context in organs, or even multi-organ coming together and have some emerging properties. A lot of immunological disease are that way. You cannot do exhaustive high-throughput experimentation in animal systems or in organs or in all of these complex translational models.

You can do some experiments, and these are expensive and high stake. The ability to build a model that can be trained on massive data where it is possible the scale and be training in a way that can be fine-tuned and transferred to make high-quality causal predictions in these complex models so that we can go into the lab and have the highest quality hypothesis possible to validate. I think that's the whole point about building a virtual cell model. But from AI side, I think you're absolutely right that I believe the future virtual cell model should be able to incorporate multiple modalities, not just RNA expressions.

Spatial single cell RNCIC is already a popular technology. Even for SGBPT, we actually have an extended version. We call it SGPT Spatial that is specifically designed for spatial single-celleromics. And we also have papers on early attempts to try to take the H&E images, trying to predict the gene expressions. There's already some signals you can find. So eventually what I predict is that, a virtual cell model will be able to integrate not only RNC can integrate more functionally

related, for example, proteinomics or other regulatory side of omics such as ATACIC to overall combine all your descriptive omics data sets to predict the future states of the cellular functions. I think that's probably the future for virtual cell model. We had Ron Alpha and Dan Baer from Noetic as guest recently, and viewers who want to hear a little bit more about that, I think we go quite in depth there. So if you want to get back. So background, you can go to that. But can you explain a little bit about what spatial transcriptomics and spatial proteomics are? So maybe a bit of a history lesson here. Before we had singles RNASEC, we had RNASEC, and before that we have microarray technologies. What RNA-Syka microarray used to do is take a chunk of my tissue, grind it all up, put in a blender, imagine make a smoothie out of it, and take all of the RNA from different cells in that piece of tissue

and measure all of their expression levels. It is great. For the first time, you can measure gene expression all 20,000 at the time. We used to do them. We used to have to do them one at the time. But it is not great in that we don't know which RNA came from which cell. And this is particularly a problem if you're dealing with a multicellular piece of tissue. You want to attribute RNA to the immune cell, to the skin cell, to the fibrobloss, to the keratinocytes, but you can't because you've gone everything up in a smoothie.

What single cell technology allowed you to do is analyze them cell by cell. So now I can attribute RNA gene expression to the cell that they originate from. But there's still a problem. I don't know spatially where these signal come from. And for many disease, it matters, right? In immunoncology, for example, you want to know when T cells are close to a tumor cells or when a T cell is not able to penetrate the solid tumor, what is the difference between them? Or when a T cell is attacking the tumor cell, when a T cell is not, what is the difference about that?

And for that, you need spatial information. You need to observe cell in situ in their context. And so now there are different technologies that solve that. problem. Essentially, take that chunk of tissue, I don't have to grind it up anymore. I just make a cross-section, lay it down in a piece of slide, and I can measure its morphology using standard techniques like H&E staining. I can then also measure many protein expression using multiplex iF assays, immemophorescence assays. Ultimately, I can also look at the gene expression up to genome wide in all of these cells in their native spatial coordinates.

by using some of the latest spatialomics assays. So you have the X, Y, coordinates of every cell, but also all of the molecular analyze that we talked about earlier. And that's an exciting new direction for genomics field. I can imagine that the spatialomics adds more difficulty to AI modeling. Because instead of looking at the individual cells, you have to look at the neighboring niche cells to better kind of learn the representation that is,

spatially cohesive, that is the challenge the current spatial foundation model are facing. But that context is going to be crucial for, I mean, understanding, let's say, cancer, where the interaction of immune cells and cancer cells and non-immune or cancer cells. Yeah, that is absolutely crucial for getting a disease biology. You can extract spatial-aware biomarkers to predict some of the clinical response. I think that would be extremely important to build such models. Getting back to Excel, this, you know, presumably can inform a spatial model as well, right? Because you can, you have like one cell in one place.

If you can imagine, okay, I can just throw away the coordinates and just do inferences on one cell at a time. And now I can create a more complicated model that does that, but it also knows who its neighbors are. You're absolutely right. But the current version we were releasing, we're not dealing with spatialomics. Definitely our ongoing work and the next version of Excel will be able to infer the spatially aware representations for different sales. I see. We've talked about the data collection a bit.

Let's talk about the architecture, get some red meat for the AI engineers listening in. Sure. Let's get to the history of virtual cell modeling, particularly virtual sale 2.0. I think our SCB kind of sets the foundation for most of foundation. models of single cells is that we adopted kind of auto-regressive training, extremely similar to how ChargbT is trained on languages, right? We use its next word, next token predictions. So we mimics the way how Chachabit is trained on languages to change the single-cell foundation model on cells. By doing that, they have to assume an inherent order of genes. The way we assume the order

of genes is by attention mechanism. There's many other methods that are using different orders of genes, some as simple as just rank the genes based on the expression values. There's also more complicated kind of methods to rank different genes. But in hand, you have to assume an order of genes. And just to be clear, so when you talk about genes, those are intrinsically ordered, right? They're a sentence spelled out an ATGC, right? So genes themselves have the nucleotides and there's this long chain and that makes a lot of sense to have an order to them.

But what we're talking about is something different. That's the expression data, the expression levels. So the expression level means how it's just a count for each gene of how many of these genes did I see when I was measuring. For the sequences, the order of ATG make total sense to us, right? But for expression data, they're literally just matrices. So it's really hard to assume an inherent order of genes. You might if we shuffle the order of genes. I think the biology doesn't change much.

However, because of the way language model is trend, everybody has kind of preset tricks to train such a model. So it's easy to adopt. That's how all the foundation model are started for single cells. And then I quickly realized that with diffusion language models, we actually don't need to assume the order of genes. Instead, we can have a bi-direction or diffusion process to generate such high-dimensional gene expression datasets.

So just to think about it, what's the difference between auto-regressive training versus diffusion language models is that you can think of auto-regressive training as typing. For example, I like coffee. You have to type I and then like and coffee. There's inherent orders. But Diffinian language model, you can treat it as editing.

You iteratively generate a sentence from a very vague, very rough sentence, and then you can iteratively refine it. So same thing with gene expressions. You can generate a very rough representation of the gene expressions, and then iteratively from noisy representation to more refined representations. So you kind of iteratively edit the gene expression predictions until it minimize the losses. So this is a very different philosophy to generatively predict the response after perturbation. It turns out it actually fits more to a single cell iron sick.

So that's why we switch it from SDGB-like model to the current X-Sail model, which are using diffusion language models. When I think of transformers, they're fundamentally objects which operate on sets. The community spends a lot of time trying to make them things which have some sort of causal ordering to them. but if you just naively take your transformer, it's a set operation, right? So given that, why think about this in terms of diffusion or, you know, autaggressive LLMs, why not have your initial prediction strategy be something like take a, just a set of genes, each of which has its own kind of one hot and coded identity and then use that as sort of a prediction?

That seems like a much more natural architecture to me. And it's not just your work. A lot of people work on things like this. And I have been somewhat confused why there's this bias in the community about this. So I think what you were referring to is more related to representation learning, where you can take sets of genes and try to project to low dimensional latency space. But what we care about for building generating, generating for virtual cell, because you want to predict the dynamics of sales.

So we want to have a generative model. So that's why we're mostly using decoder-only architectures in order to generate the full transgatomic instead of just predict a predefined small set of genes because you want to model the whole gene gene regulatory networks, which are extremely kind of high dimensional, right? So just to be clear, input is genes plus a perturbation output is new gene expression levels, is that gene expression levels plus perturbation is input output is times sales. Oh, like for each cell?

Correct. Correct. That is correct. Right. Okay. And so the way that I think about, the way I think about diffusion language models, and you can correct me here because I don't know a lot about them. But the way I think about them is they're like Burt, but you do it over and over again.

Is that, is that kind of a good thing about it? Yeah, that is a rough understanding of how diffusion language model works. Yeah. So you just apply the diffusion, the diffusion process is. like basically unmasking or editing once over and over again. The similar to how like an image diffusion model kind of refines the image over and over again. In this case, I'm using BIR. So it is a transformer basically.

It is a transformer architecture. But it's like repeatedly updating the sort of sentence in this case, which is a bunch of expression levels over and over again. That is correct. Actually in our paper we show that, As the number of diffusion steps goes on, the loss function keep decreasing the fitness of the prediction to the ground shoots keep increasing. So this, which means the model starts to understand how iteratively refine the predictions. I see.

We're talking about diffusion versus auto-regressive. I noticed in the paper there's a bunch of discussion of preconditioning using a whole bunch of stuff. You want to talk a little bit about that? Another major innovation we made in Excel is the way we incorporate prior knowledge into the model. So incorporating biological priors has always been a good idea in biology in general because biologists spend decades to understand some of the biologists already. How do we tell the model some of the prior knowledge, some metadata about the cells? Before Excel, what people do is they try to increase.

incorporate a single type of priors. For example, gear using gene-recognit network as a prior to predict the prohibitions. Sedgebd sometimes trying to incorporate PBI as a prior as well. Excel, to my knowledge, is one of the first models I'm trying to incorporate an extremely diverse sets of biological priors. So in our preprint, we incorporate five types of priors,

including literatures. as simple as just ask CHHGBT, tell me everything about this gene. And then we embed the output as the embedding. So it's like gene PT. Exactly, that's a gene PT. And we also incorporate PPI, protein protein interaction networks. We also incorporate the DEMAP, which is cancer-related essential gene information, morphology information. We even trying to incorporate

SGBT embeddings, which is basically cell types. So with a set of prior knowledge as conditions to the model, the model start to have more accuracy in terms of context specific predictions. And what's more interesting to us is that by looking at the weights of different priors, we can actually understand which prior knowledge are more important to this particular cell types. So it adds more interpretability to the models. So we find that combining diffusion language model, plus a very diverse set of prior knowledge is.

Excel does much better in generalizing two unseen contexts. So this is some of the AI innovations we made for Excel. Do you now need to provide all of that context in order for the model to work? Or those are preconditioning that it can also do without if you want? We don't need to incorporate these prior knowledge anymore because these are already learnable parameters inside the models. However, what you suggest is more promptable or in-context learning for virtual cells. We can do that as well, basically by adding more conditions into the prior knowledgees

so that to prompt the model to predict towards certain directions. In other words, the model now takes advantage of the learning using the priors that you provided during training and doesn't need them, but it has some advantage because you provide them during training. But you can even get more advantage if you are able to. to provide those priors during inference. Yeah. Okay. Wow.

Nice. How much does that matter? I mean, whenever I see big machine learning papers with tons of things thrown in, I'm always wondering, where's like the big alpha and where's the little alpha? How much are, you know, is this just some, are these adding this a little bit of incremental performance boost or, I mean, are all these actually crucial to general generalization? So there's multiple factors we have to consider. How much contribution of data contributed?

how much of the contribution, the AI architectures contribute. Even for the architecture, what's the delta from switching to auto-regressive training to diffusion. What's the data from the prior knowledge? Certainly, all of these needs very specific ablation studies. From empirical experiments, we find that the qualities, the amount of the data sets matter the most. This is why we were extremely excited to publish the PISA's data sets, which has 16 different cell types and across 25 million cells, and it's genome-wide. It has kind of a huge tensor if you really think about from computer perspective.

Genome-wide perturbation, genome-wide transcript comics, plus number of cells, plus a number of conditions. So it's a massive tensors. And because of the poor screening technology, we don't have batch effects. So you don't need the model to climb the heel of batch effect. That's already advantage. So we find that trend on perturbation datasets, high-quality perturbation data sets, already gives a big boost to the models.

We also did ablation that if we train all the virtual cell models out there, including state cell to send us original SDVD on the same data, sets, what's the data we are observing? We report the results there as well. We find that switching from auto-rogressive training to division language models give a significant improvements over some of the harder tasks, particularly generalized to unseen tasks. And there's a prior knowledge, more or less condition specific. For certain cell types, some of the prior knowledge make a huge difference, but for the prior knowledge

certain cell types, the delta seems to be marginal. We are thinking about how to better incorporate the prior knowledge. We still believe that let the model know a big chunk of existing biology should be helpful, but maybe it's the way we incorporate the product to cross-attention, limited the scope of the metadata, but I think it's certainly a research topic. But overall, if we have to give an order, my order would be the quality among scale of the datasets and then the architecture and then the prior knowledge. But certainly, this only applies to our Excel.

I'm sure there's different choices of architecture and have different ranks of contributions. First of all, this is really fascinating, very cool model. I hope everyone has a chance to look at the paper. There's a lot of, obviously a lot of resources that were put into. doing this. I don't know if you guys can disclose how much. It's a lot of money, whatever it was. Operating wet lab, probably very complicated training runs. I think there's a couple four billion parameter model, is that right? Four point nine. Yeah, four five billion parameter model. So much larger model probably took a lot of GPUs to train. What's the lift that you get from this effort

versus let's just put the money into like wet lab work and the sort of traditional. additional pipeline that basically has been the status quo up until now. Biology is a multi-scale discipline. There are cells, there are DNA sequences on the most fundamental level. There are cells. There are multicellular pieces of tissues, co-cultures. You have tissues. You have animal systems.

And finally, you have human. I think we would like to be able to do causal protection towards the right of this spectrum, ultimately do cause a prediction in human, know what drugs will work in which patients, but that's very difficult to collect high to put data on. And so the whole vision of virtual cell is to generate data where it is possible so that we can transfer the causality prediction towards the right, towards the more translational, the more complex systems. Certainly, you can mine the data already. We generated a lot of data, as both said, seven screen, 16 different and biological context, genome-scale perturbation.

There's a lot of good ideas in that already. There is a figure that we put out in the preprint that just look into inactivation of T cells, we already saw some, you know, we saw known biology, TCR complex. We also saw some palliative new biology, which were very excited to validate in the lab. Some of that were actually also caught out in a very recent screen last December published from Alex Marsen Lab, also in the Bay Area. So very excited to see that. But the hope is to not just mine the existing data.

The hope is that the model can generalize and we will be able to do in silico experiment into the future. Nobody knows before how much data and what kind of data are needed to do that. With the whole field waiting for the demonstration, that the model can beat linear baseline in perturbation prediction, and it can generalize out of context, not just within a cell line you have training data on, but out of that context. That's why you need a model.

So what's very exciting for us is that in this preprint, we saw that generalization capability. A few demonstrations. We first did in T cells. We actually generated the data expressly for this purpose. We generated the resting T cell perturbation screen. So these are T cells in their baseline condition, not activated. And then we have a activated T-cell perturpsic. So just T-cell activation means I'm going, I'm trying to kill something.

No, these are regulatory T-cells, but yes, we activate their receptors so that they're starting to proliferate. Okay. They become more active. They can do their physiological job. And we only, critically, we only train the model on the resting T-cell. And we told the model, hey, this is how the active T-cell. look like now go and predict what all of the perturbation are going to do in this active T cell.

And the model have not seen how perturbation working active T cells. And we set up a couple of rigorous tests. One, linear baseline. Took the perturbational delta in the resting case, just transpose that linearly onto the active T cell, and that's our linear baseline. Essentially, think about this as a combinatoral

perturbation prediction problem. One of the perturbation is activation of the cell. The other is all of the genomic perturbations. Can I just linearly add the two effects together? That would be a linear baseline. And second, we apply other models from the field. And last, but we applied X-cell. Critically, X-L has not seen active cell, T-cells.

And it's able to make accurate prediction, not only on the known biology, the TCR complex, predicting their effect accurately, that these are going to inactive the T-cells, which is exactly what we would expect to see, But also, it predicted the punitive T-cell inactivators that we found in the screen correctly as well. So that's very exciting to us.

And that suggests the possibility that we might be able to use these virtual cell models completely out of context in the unseen context and predict new biology. And so we're very excited to follow up on those heads and validate them in the lab. Just very briefly, a couple other cases that we saw exciting generalizing capability of this model. Remember, we did a multi-cell type

differentiated IPSC experiment, there we specifically held out one cell type from training. So the model has not seen that cell type. Train on the other cell types as well as the rest of the data sets, the model made very good prediction across thousands of genes, thousands of perturbations in that unseen cell type. So again, suggests the model's ability to generalize out of cell type. And the last experiment, I think we're very excited is that we train this on T-cell cell line. but there was just very recently a primary T-cell proterpsic published from Alex Marston's lab. That's an impressive amount of work.

It's not easy to do this scale screening in primary cells. Very few labs have that kind of capabilities. Much easier to do that in T-cell lines. Again, the model is able to generalize out of cell lines into primary cells and make accurate predictions there. So they actually perturbed primary cells, not cell lines. Primary T cells harvested from donors. And we were able for multiple donors.

And Xcel train on just one T cell line is able to make predictions across multiple donors from primary T cell experiments. This is a validation of the whole theory, right? That you can train on these slightly weird cells and that it will be good because you know you're covering the domain well enough or whatever it is, that you're able to actually predict in real cells that come directly from real people. That's right. I think building virtual cell is not to replace biological experiments,

as you mentioned. What we're trying to do, really the holy grail of virtual cell is to have a model to generalize to unseen contacts that is harder or even impossible to conduct biological experiments on. So far, Excel focusing on cell lines. And eventually we want to extend to more complicated, biological systems such as animals, organoid, and eventually, as we mentioned before, to patients, to human biology, right? And bearing a few numbers, 90% of disease has no cure,

and most of the drug failed at the phase three clinical trials on patients. And the success rate of phase three trials is as low as 5% to 10%. And phase three means the final stage on the patient trials. So that's when you generalize from toxicity in phase two to efficacy in phase three. From a small cohort. From a small cohort. Oh sorry. Toxicity is one. First one.

Two is small cohort. And so a lot of large cohort. So the generalization problem, okay, this drug, I've very carefully selected my patients and it works pretty well. And now I get a bunch more patients and suddenly it doesn't work very well. And that's the big problem that you're. The promise of virtual cell is that can we build such a model that learn all the causality of biology so that can be grounded to predict the response eventually on patients so that we can, for certain drugs, we can select the right patients to conduct the clinical trial zone.

So this is a long-term vision, but we already see some early hopes that Excel trend on diverse set of causal data sets can already generalize to some unseen cell types. So certainly there's a lot of experiments to be done to validate this model, so even continuously finding this model. But I think we certainly see some early hopes. So you were talking about linear models, and this brings up this famous or infamous arc challenge about, you know, perturbation. And there's been this theme about complicated foundation models, oftentimes not beating linear

baselines. And I'd like to get your take about that as terms of what is this, you know, first of all, is this different? I mean, I think some of maybe your own models might also have had trouble beating linear baselines in the past. Is there something different about your current data strategy or where you're going and where's the field going? And what is the role of foundation models versus, you know, these simple baselines? Yeah. A few. A few. things. First of all, those benchmarks, as you mentioned, are conducted on repro local datasets, which are very small datasets. And the metrics people report are mostly M-A-E. Certainly, you can imagine if the, because single-cell data set are so sparse, the average

profiles of all the cells, certainly you can imagine is a great minimum, kind of local optimum to minimize the MAAs. This is why sometimes the average profile of cells has lower MAs even than technical replicates, which are considered of ground choose for kind of perturbation experiments. So that itself shows that that metric is not reliable. However, most of these benchmarks are still comparing foundation models that trend on static expression data such as cell by genes.

SGB is often benchmarked against. In turn, we also find that when it comes to MAA, sometimes SGB kind of failed to outperform linear models just because of the reasons I just stated. And what sets Excel different from these static expression models, such as SGBO gene formers, is that we actually, instead of trend on gene expression datasets, we train on causal data sets.

We train on massive amount of genome-wide perturbation data sets so that it learns better about the data. dynamics of the interventions. And in our preprint, we extensively compare with linear model as well. And as you mentioned, linear model totally failed to extend to unseen cell types. You can quickly imagine why. And I believe that foundation model or other more complicated AI models that trend on the right data will outperform these linear models in harder tasks, particularly in generalization

tasks. And that's why I keep mentioning the right data set with the right AI model will lead to huge improvements. But I think the field still needs to see more biological validations to be more convincing that the virtual cell direction is the right one. Yeah, I think the field suffered from a lack of consistent and uniformly accepted benchmarks. What gets measured will get improved. And in our paper, we measured, I think one of the metrics that we put a lot of thought into and saw the model really shine is metrics around gene expression changes. So, you know, Pierce and Delta, the similarity between predicted changes and ground truth changes upon the perturbation. That's very hard to cheat on.

You have to really get the changes right. And what really blew my mind away is when I saw the model make predictions, just print out the heat map of the genusversion changes. Look at the actual raw data and line up the linear baseline prediction, the ground truth and the XL prediction altogether. It's visually very clear to see that XRat prediction is very much more similar to ground truth

than the linear baseline. This is a wow moment I was talking about in the beginning. That's right. And it's not hard to understand why. So this is the first time that someone can put together not just one perturbsic, but seven genome biperchirpsi campaigns together. Something that jumped out to us, biologists, right away,

is that some of the perturbations are context universal, meaning that the genes do the same thing in regards of the cell types you experiment in. Might not be surprising to you that these are your housekeeping genes, right? Of course, they do the same thing in every cell. And then there are all of these other clusters of genes that have very context-specific functions. they do different things in different cells. Again, not hard to imagine why. In IPSCs in our stem cells,

we saw developmentally relevant genes, genes that are important for neuronal differentiation. They only light up in IPSC experiments, of course. Right, that makes sense. So think about biology, it's so complicated. You have to capture these context universal perturbation effects. You also have to somehow learn the context-dependent perturbation effects. It's not hard to then see why a very sophisticated, non-linear

model is able to capture and learn all of those biology much better. Are your perturbations always single gene perturbations or do you have more? Because my understanding of regulatory networks is oftentimes sometimes it can be a single gene does a ton of things. For example, I think males are just differentiated due to one gene being enabled at like day seven of embryo development or something and that differentiates everything. It's this one gene. But then sometimes you have large networks of genes, which all are very redundant, which allows

for more subtle feedback mechanisms and so on. So I could imagine a lot of single gene perturbations as being kind of irrelevant and that you might want to start having a more commentatorial strategy here. Yeah, that's a great question. And Bo and I have thought about this a lot. Actually, it's interesting that you brought up reproductive biology. I study a lot of in female cells, the dosage compensation mechanisms. There, female cells have two X chromosomes.

Male cells have one X chromosome to match the dosage output from X chromosomes. Strategy that the mammalian cells employ is one gene that produce a RNA. It does not encode for any protein, just a non-coding RNA. That RNA wraps around one of the female X chromosomes and turns down, most of the gene expression from that chromosome, shove it away in a corner of the nucleus and it's called a bar body. It has never heard from again. And so absolutely agree with you.

One gene can do a lot. But in biology, you also have redundancy. You have compensation. You have all kinds of mechanisms where knocking down one gene is not sufficient to always see a phenotype. What if four genes redundantly do the same thing, right? Taking out one is not going to be sufficient. So where we started with one cell type, at the time, loss of function, single gene perturbation, and look at only RNA expression

as the output. We're expanding the platform along all of those axes. So that's what we do today to build a scaffold of the data for training models like Excel. We are now beginning to grow in all three axes of the platform, going beyond Transcopatomics alone to like a multimodal data, going beyond just one gene perturbation alone to look at all also pathway activation and inactivations, turning on and off entire cascade of gene,

chain reactions. And also going beyond just cell lines, monocultures into more and more complex, translational relevant systems into primary cells, into organoids, and doing even direct in vivo perturbation swings. So we believe with all of that expansion, the data will be all the more exciting to train models on.

This is also why we incorporate PPI networks as the prior knowledge into our model. And although the model right now are trend on single gene perturbations, but once the model is trend, you can actually predict combinatorial perturbations just on the model in only silicon, right? So in the sense that you can just perturb the tokens

of two genes at the same time and see what's the response. Certainly, without training the actual combinatoral prohibition data sets, the accuracy may not be there, but at least with the existence of such a model, those, we can start to generate hypotheses using incital perturbations. How do you see the role of the scientists changing in the age of AI? And you may, because you're not using, your focus is not language models themselves in

agentic science and things like that, then you may have a different, slightly different take. You're building, you know, these very specific models. But still, one, how are you and your students able to maintain something? such a high pace. And I suspect it may have something to do it generated by AI partly, but also and how do you see the role of the scientists, the academic, changing? Yeah, that's a great question. So my official split of time is 80% on Zara, 20% on my university affiliations. But turns out the reality is 100% on the era, 100% on this. You invented a time machine. That's the answer. So the way I trying to keep up is,

my lab use lots of agentic AI trying to monitor all the AI papers every day. And every week we have lab meetings. We're trying to discuss different topics in AI for biology, AI for healthcare, et cetera. And to be very frank, even as a professor, I find it's extremely hard to catch up. The pace of AI is just so important. incredibly fast. And to the point that sometimes I feel anxiety, waking up says, oh my God,

these people are already so many people published and what happened to our existing unpublished work. And certainly you can imagine students probably face 10x anxiety. So sometimes I try to encourage students to really kind of using different tools trying to stay focused. You know, finding the niche area that we become an expert. on, right? But with the era of generative AI, now agentic AI, I find the way people, at least academics does science, are extremely different now. Overall, most of professors or students in academics start to be very struggling in terms of fundings, in terms of the pace of publications. And that's why you can see lots of major breakthroughs come from industry, right, like AlphaFo.

for example. So how academics survive or even thrive at such era of agentic AI is certainly something everybody is thinking about. We see lots of faculties left university and join industry for simply for the reasons of resources, right? If you are doing research on AI, do you have enough GPUs at your school? It's the first question you should ask When a student join a professor's lab, the first question they often ask is how many GPUs do you have? So certainly in that sense, industry has a major advantage over academic labs. But I think what academics labs have advantages on is really the pace of innovations and also the niche areas of this specific academic lab can be extremely expert on.

So also having the freedom of thinking sometimes also make you easier to innovate on ideas that maybe industry people didn't even think about. Overall, I think the whole field needs to be a lot more innovative to catch up. And hopefully with the help of different tools and hopefully the government starts to invest more into academic because I still deeply believe that academic is the main sources. of innovation for the whole field, and particularly when it comes to biotech. So hopefully we see more investment into academia so that we stay afloat. I agree with you, but why? Why do you think that it, why shouldn't money just go to industry? Why should the government put any money into academia? I still believe the power of, you know, academic freedom.

This is actually the original motivation we have, such, the existence. of academic professors who not only we teach, but also we do research. And there's also benefits of teaching and research at the same time in the sense that when you teach a subject, you actually have to become the experts on and that force you to keep updating your knowledge base and then find easy ways to convey your knowledge to students. And by doing that, you actually start to innovate on different ideas and my For example, I teach a big class in University of Toronto about deep learning and neural networks. It's a gigantic class with 600 students every year.

And by teaching that, it forces me to update the slides, lectures every year. And myself, reading lots of materials trying to update myself so that I can find ways to convey some of the knowledge into students. So that's how I keep updating myself whenever there's. I remember vividly, we have GP2, we updated the lecture, then different language models, how the multi-threat GPU communication is used in training, largest scale neural networks, etc. So I forced myself just updated the models. And by talking to different students, we really generate lots of novel ideas to apply the cutting-edge AI models to very specific niche areas in biology or in healthcare.

Maybe it's unique to Canadian academic system. By being a professor in academic, we also have access to lots of healthcare data sets, which are very hard for industry to access due to many illegal reasons or regulatory reasons. And that's why you see some of the papers we publish through academic hospitals in Canada, where we develop some of the state of arts

foundation model for ultrasound images. So that's why we're not. what I mean that there's a certain level of freedom of academic thinking that really drives lots of innovations and I still believe maybe it's biased but myself still believe that having certain level of freedom of academic thinking will leads to lots of kind of innovations that is unsinkable in industries. I 100% agree with a bull there. So, you know, just thinking about the lab, workflow that we do. A lot of these are building upon

innovations that were first pioneered in academia as well. CRISPR, of course, was discovering academia. CRISPR applied to mammalian high-thupert screening also demonstrating academia at first. Singles RNA-SEC, this kind of drop-up encapsulated singles RNA-SEC first demonstrating academia, then different companies tried to build it up into commercial offerings.

Putting all of these together to do perturbs-SEC at first, also pioneered in academia, right? Chris Box lab, Jonathan Weissman's lab, FVGELF's lab, many of the pioneers in academia. And then I think these, especially on the lab side, these innovation takes so long and the discovery process can be so accidental, right, that it's perhaps not ideal for pure industry to take on.

But once they show early promise, scaling them and robustifying them, and generating data that's not only massive, high quality, especially for AI scale, I think that's something that can be very well done in industry, both the mindset, as well as the kind of the resources that we can support. That sometimes can be hard for academia labs to match. Zara has been very generous with releasing your data sets and your models. You've been, seem to be very committed to open science.

Given some of the things you were just talking about and, you know, the discussion about, like, what is the best, most important data strategies for virtual cells or understanding human biology, where do you think academia should go next now and next? Since, you know, Zara has probably a budget comparable to, you know, probably dozens or hundreds of bio labs right now, what do you think that if you are an academic, a professor, especially in a wet lab, what would you want to be focusing on? First of all, myself is a deep believer of open science. That's why all the models we talk about here that are open source.

You can find all the data weights from my lab GitHub. It's very thorough. Yeah, thank you. And the reason I believe open science is that, as you mentioned, most of the time academic lab, we started an idea and we prototype it. It's not scalable. It's not even a good product. and industry can take it to scale things up.

This is also why SGBD quickly become one of the most widely used single-cell foundation model in former companies. So that's very encouraging to us. And this is why Zara is also start to open source some of the datasets, some of the models. Part of the reason is that we believe virtual cell is such an early field, and it doesn't help to withhold certain datasets or certain models, because it's so early, a better win-win situation is everybody gets to in this field, start to contribute data together, start to contribute models together to exchange ideas so that this field can move forward in a much faster pace.

We see successful examples in protein space, right, because of the availability of open-source data in PDBs, therefore we have models such as AlphaFold, RosettaFold. Again, AlphaFold to also open-source, therefore we can quickly iterate different models. That's why you see kind of a booming situation in protein space. We want to do the same thing for virtual cells. Let's put all the data set together. Let's have the same standard protocols to generate high-quality data sets. Let's put all the resources together to generate next generation of virtual cell models. When it comes to academic labs, St. Nguyen Chu can comment on wet lab, how wet-lap academics can survive. But from dry lab perspective, I do encourage all the, you

know, dry lab AI researchers in universities start to collaborate with industries so that they can get more resources to develop their own ideas. And with the era of agentic AI, now everybody can code. So it's more important to have a right taste about your project so that you don't just let, just burn tokens without purpose, right? So we want more academic students, academic professors. is to have higher taste of research so that we make the right utility of agenting AIs. As a professor, your job is to have taste, obviously.

But as a student, how do you develop taste in a world where so much of the thinking scientific process could be essentially outsourced to an LOM or, you know, there's not a world where you're forced to bang your head against something and learn taste by the hard way? This is why we need academic training where you get into a field you know nothing about. And hopefully after you graduate, become the expert about this particular topic in the world, right? This is why you have to go through different programs to talk to your peers, talk to your professors, to get an idea about what's a good research taste to begin with. But more importantly, I always teach students that the best way to learn something is to just cook

program it. By programming, you kind of know what's the details hidden in all the mathematical equations in the paper, which often you admit. But in the era of agentic AI, since a slight different in the sense that we used to spend lots of time coding, a little bit of time just debugging. But now we let the agent do most of the coding, but we spend most of time debugging, which seems to be definitely interesting to me. And we had lots of discussion with students in the lab, what's the best way to spot bugs by AI's? So how do you find places where AI is particularly good at?

Also, how do you find places AI are still limited at? It's kind of you need lots of China era as well. And in the end, you still need to kind of validate your model using real world evidence, right? So that's why collaboration with wet labs, collaboration with clinical teams to validate your model, provide a feedback signals to your taste, as we discussed, is a very important training program. On the wet lab side, academic has extremely important roles to play. I think we'll enter a field of an era of symbiotic innovation and cross-pollination of ideas. Just like in AI field, I think we have great ideas coming out of academia still all the time.

But industry now increasingly are contributing new ideas on architecture, on all of that as well. In the web-lap side, certainly industry seems to be able to scale these type of data generation quite well. But biology is so much more than just cell business. based perturpsic. Beyond RNA seek, we would like to measure many other analytes, right? Proteins, metabolites, lipids, protein-protein interactions. How do we do that at scale? Beyond individual cells, we would like to be able to measure cell interaction, spatial, cell in their native context, or even whole animal level in vivo perturbations. Again, how do we do that at scale with lots of great innovation coming out of academia?

Actually, just one great paper last week. And so how do we connect all of these together? I think we have many years of work ahead of us to fully crack data generation for all biology. And that, I think we need the scale, the industrialization, the innovation from industry. We also need that from academia. I think we'll together move this field to the next level. One question that we've been trying to ask everyone is in your field, which you could say maybe is AI and, you know, sort of high-throughput experimentation or however you wanted to find that.

If you could wave your magic wand and have a bottleneck removed for you or a key problem solved, what would that be? Protein. Okay. If there is a way to do protein sequencing or high-thuper protein measurement, the same scale that we can do genomics, that will be amazing. I'm training in genomics field, but if I can do that, I would incorporate that technology in a heart speed. RNA is amazing, it foreshadows which proteins are going to get made, but protein by and large

are the functional units in the cell. Not only does their abundance matter, their post-translictional modification matter, their localization in the cell matter. If you can measure all of those things, their conformational states, their modifications, their abundances, their localizations at scale, single cell or even spatially, I think such data this as would be incredibly useful to train the next generation of financial models. I know there's a lot of innovations in that direction. Can we just see that become come of age?

My hope is, I hope to see a breakthrough in sequencing technology, not just the reduced cost, but sequencing technology that can sequence the same cells at different time points. I think this is much lacking right now because in order to sequence the cell, you have to kill the cell. So can we have a technology that can measure the cell states at different time points for the same set of cells? I think that will bring a very different dimension to the data set so that we can start to measure the temporal dynamics of cells. So far, everything we measure, everything we model is extremely static. So can we have a technology that measure different cell response at different time points for the same set of. cells will unlock massive opportunity to model the dynamics of cells.

To me, that is a real virtual cell. That's a really interesting idea. I never would have thought about that. Wow. Would you be okay with even just partial, like small snippets of genes or maybe three prime regions of a small number of transcripts or something? Yeah, we can start with a small set of gene panels to begin with, right? But eventually, if since we're talking about the magic,

Eventually, if we can have a system that observe how cell evolve at different time points and we have enough data to actually model such development, I think that would be real virtual cell model. There's a small attempts in, oh, sorry, earlier attempts in just, for example, sucking out portions of the cells, taking almost biopsies from the cell to do, you know, a fraction of the saloplasm measurements. So that might be similar to the idea you talk about. There's also work from Poplinis lab to have the cells secrete little vesicles.

And you harvest that in the cell culture media to measure what the cells are producing longitudinally. But there hasn't been technology that can let you measure the entire cells from while still keeping the cell. You can't have the cake and eat it. Cool. Yeah. Thank you for taking the time to chat with us. It's been great. Learned a lot.

It was a lot of really interesting discussions. And I think, and I especially really appreciate your commitment to open science and all of the cool models and data you've released. Is there any last, like, thoughts you have or anything you'd like the audience to know, follow up with? Overall, I think virtual sale is such a new and fast-moving field. We hope to have more and more people join us. And our Excel paper is out. And we look forward to receiving your comments and feedbacks. And also, we're hiring.

Yeah, we are always looking for talented engineers, technologists, biologists, drug hunters, AI, scientists, computational biologists. So look on zara.com, look for the open roles. We'll be happy to chat with you. Thank you. Thank you very much. Great. Thank you.