No Priors: Artificial Intelligence | Technology | Startups
00:26:26 · Sarah Guo, Priscilla Chan, Alex Rives
[00:26:26] S3: Whether the timeline is ten years, hopefully less than one hundred now, I think this is useful for maybe your average doctor or patient, human being, everybody's a patient, to think about like what's externally visible in the progress here. You worked with patients for a long time at UCSF. Like, what should doctors look out for? What should people look out for if you're actually accelerating progress? [00:26:52] S1: This is the part You know, I'm super excited about the progress, especially with this launch that Alex and his team have put forth. And I think it's very clear that science is gonna start moving pretty quickly. And I think the thing that's less clear to me is exactly how we translate to the clinic and what that looks like. And I think what has to change is actually the way we do clinical research. And my hope is that we're really shortening the distance between bench research and patient impact. But there's a lot of steps there that we need people who actually take care of patients to think creatively and think about how to deploy safely. And that's a gap that we have some work in. We partner with Jennifer Downey on our CRISPR Cures program at UCSF. So we're dipping our toe in understanding how the deployment of research needs to change given how quickly research will be progressing. But that one is still, I think, still shaping up. [00:28:04] S2: Maybe I could say something about our most recent launch [00:28:07] S3: because I [00:28:07] S2: think it [00:28:08] S3: also Oh, yeah, kind please. We should ask you explicitly about it.
00:28:11 · Alex Rives
[00:28:11] S2: Yeah. So, you know, because I guess it was just a week ago, about now, so we announced the new ESM fold. And so this is basically an open system for scientific discovery in protein biology. It's a world model of protein biology that's been trained. It's a language model based, so it's been trained on billions of protein sequences, kind of learns these emergent representations of protein biology. And then we can use it to make predictions of atomic resolution protein structure, and we can use it to and it's really fast. So it's blazing fast. So it's kind of you know, illustrating this Pareto optimal frontier of kind of speed and accuracy in structure prediction. And so this allows us to kind of characterize, you know, really vast kind of stretches of the protein universe. So we folded over 1,100,000,000 proteins and predicted their structures and identified kind of features connecting all of them through mechanistic interpretability. But I think the thing that I thought was most exciting about this model is this really general model of kind of protein biology. And so you can use it as a world model. You can actually really start to search the space of the world model to design new proteins. And it's really hitting state of the art across pretty much every structure prediction benchmark, and especially on protein protein interactions and protein antibody interactions, which is really critical for therapeutic design. And so what we found is you can actually now use the model to design proteins and to design actually single chain antibodies. And so you can do all of this digitally and then, you know, really in a small number of experimental trials, basically like a 96 well plate, you know, select from hundreds of thousands of trajectories digitally, actually synthesize, you know, 96 proteins, test them in the lab in a really kind of short, easy experimental cycle. And we found nanomolar binders there. And so, you know, that's really the level for therapeutic activity. So it's really, I think, showing that you can have these kind of general purpose models that can you know, we didn't design a model for antibodies. We didn't design a model to, you know, to be able to bind one particular target. You know, we just designed a model that could understand proteins, and you kind of get protein design as an emergent property.
00:30:42 · Alex Rives, Priscilla Chan
And then I also think it illustrates this kind of the power of open science and open source because, you know, we release this as basically an open discovery engine. And so really anyone can build on it. And so it takes what are these really intensive laboratory experiments where, you know, you have to screen through hundreds of thousands or millions of antibodies and high throughput screens in the lab. And, you know, you can really just kind of spin up an instance and compute and now, you know, be able to generate antibodies. [00:31:17] S1: You should say more about sort of like we took that data when we looked at an antibody screen, and then we validated we looked at PDL in cells. And then we looked at it under the cryo EM and sort of how all that complemented validated what you were seeing in the models. [00:31:34] S2: That's right. Yeah. So I mean, I think it's really critical, you know, to actually go and characterize these molecules in the lab. It's you know, we have a structural biology center here. We have incredibly powerful cryo EM microscopes. And so we're really able to kind of look at these proteins biophysically and functionally. And so, you know, we designed proteins for several therapeutically relevant targets, and we're able to confirm their function and It's [00:32:05] S1: some pretty light when it works the way it's supposed to. [00:32:08] S2: Yeah, it's very amazing. We're to look at the structure So you can see atomic resolution kind of at the binding interfaces. Correct.
00:32:08 · Alex Rives, Elad Gil, Priscilla Chan
[00:32:08] S2: Yeah, it's very amazing. We're to look at the structure So you can see atomic resolution kind of at the binding interfaces. Correct. [00:32:15] S4: I know a lot of your work is really focused on basic research and kind of building out the fundamentals. If I look at actual translation into drugs or drug development, often a clinical trial will be fifteen years. It'll cost $1,500,000,000 About 50,000,000 of that often is the molecule in preclinical work, and it's a few years of work. And then the other 1,450,000,000 and decade plus is actually the drug development side of it. A lot of that seems to be gated on some regulatory issues, some of it's recruitment, it's a variety of things. But a lot of it also has to do with the failure of drugs and trials around things like absorption or toxicity or things like that. Have you considered at all tackling that other chain of sort of molecular design and thinking, or is the primary focus more on the basic biology and sort of the initial sort of molecules? [00:32:56] S1: I mean, at least my hope in building this like comprehensive model of how, you know, cells work is actually also being able to predict off target effects. [00:33:06] S4: I think [00:33:06] S1: you can do some of [00:33:07] S4: that
00:33:07 · Priscilla Chan
[00:33:07] S1: actually with biological models. Because right now some of the off target effects are we just didn't know, you know, your kidney cell also expressed this receptor. And then when we test it in human, like we see it happening and we see renal toxicity. And so being in If you have a single cell atlas that looks at all the different cell types, some of which actually were not predicted before we modeled them, you can start looking at which cells actually do have receptors for the target you thought you were exclusively targeting and be able to predict some of these downstream effects before we get into the human trials. And I think that that's actually one of the more exciting applications of like a transcriptomic model to understand actually how the different cells will react when you intervene and do something. But I think when you think about delivery mechanisms and patient care, that's where you start having to be creative about when you ask like, what disease do you want to care first? There are certain diseases that will be easier to like deliver a therapeutic to, or the risk reward makes more sense. And, you know, I think we were all inspired by baby KJ, I think last year now, when the team at CHOP was able to deliver a CRISPR therapeutic to edit a mutation that he had that would have inevitably led him to significant neurotoxicity and altered his life. But we were able to, that disease was very carefully chosen because we needed to target his liver cells and if we could easily deliver a product that would work in his liver. And I think that's when the creativity, the wherewithal to choose the right applications can help us unlock the first applications.
00:35:08 · Alex Rives, Elad Gil, Sarah Guo
[00:35:08] S2: Maybe something just to add to that also, you know, because, I mean, kind of you described the conventional, you know, drug development process. Right? And I think, you know, these kind of tools have the potential to have a lot of impact on that process. But, you know, what's interesting is to really start to think about kind of the new paradigms that can open up. And, you know, what does it mean if if you can you know, the barrier to develop a drug, to design a molecule, you know, to kind of get through all of those stages is so much lower. And so you have programmable biology, and you can, really start to create a medicine for every individual patient. I think that has enormous implications for how we do drug development and what the future of medicine looks like. [00:35:52] S4: Mhmm. It'll be an exciting day when the FDA accepts like a [00:35:54] S3: virtual clinical trial for the phase one or something, or, know, it's based on some person's view of that person. Yeah. Or even short of that, like thinking about the specific like mechanisms where you see this acceleration, like I imagine if people feel like they can predict impact in kidney cells or have a stronger perspective on tox because they have this broader understanding, they'll be willing to try many more programs.