When I worked as a pediatrician, a lot of my patients’ disorders were basically unknown to science, and the thing I wanted more than anything else was a way to see what was happening inside their cells. How were genetic mutations getting expressed in different cell types? What, exactly, was breaking down? There was often no way to know, which made it all but impossible to find a good treatment.
As I wrote over the summer, we believe AI will help close these knowledge gaps in cell biology. The models we’re building will make it possible to simulate how cell networks function, see how they break, and learn how to fix them.
We’ve been building toward this moment since 2016. That was the year CZI launched our first scientific institute and started leading an effort to map every cell in the body. From there, we stood up more research institutes, built the largest datasets around human cell types, and established the first large-scale GPU cluster for biological research.
Because we did all that work, we are now the first scientific organization that can push the frontiers of AI and the frontiers of biology at the same time. This collection of resources for AI-powered biology doesn’t exist anywhere else.
Over the last several months, we’ve spent time refining our roadmap for this vision of virtual biology. Today, we’re giving it a name that reflects both the history of our organization and our vision for the future.
Our unified science teams will be known as Biohub. We’re excited to direct the vast majority of our philanthropic resources toward Biohub’s work in AI-accelerated science.
Nearly a decade ago, Mark and I announced our mission to help scientists cure or prevent all disease within the century. People used to think that was out of reach. Now it seems like AI-powered biology can get us there — maybe even faster than we imagined. So we’re going after this vision with everything we’ve got.
Welcoming our new head of science and joining forces with EvolutionaryScale
Mark and I are thrilled to welcome Alex Rives as our new head of science. Alex served as co-founder and chief scientist at EvolutionaryScale. He is a skilled, strategic leader with a track record of building successful teams and leading breakthrough research. As a scientist, he has done pioneering work advancing artificial intelligence for biology. What makes this even more exciting is that he is also bringing the experienced, world-class EvolutionaryScale team and products with him. Their renowned protein and genome models, including ESM Cambrian and ESM3, can predict protein properties, reason over their structure and function, and generate novel proteins. Scientists are already using these models to design therapeutic proteins and predict the clinical effects of genetic mutations.
With Alex’s leadership, our multidisciplinary team possesses all the core assets — from state-of-the-art technology and data generation to advanced AI modeling and significant GPU capacity — necessary to build advanced AI systems capable of unlocking decades of biological discoveries in years, if not months.
Creating a virtual immune system
As we build toward this vision of AI-accelerated science, we are also sharing several exciting in-progress updates on our work today. Our teams have begun work on a major new effort to model the vast complexity of the human immune system in silico. With the Virtual Immune System, our aim is to fundamentally change our understanding of immunology, moving the science from describing what happened in the past to deriving the rules that govern what will happen in the future. This opens the door to engineering human health. Researchers could simulate new immune therapies, learn how to reprogram dysfunctional immune cells and harness the immune system to prevent diseases before they become unstoppable.
This project will draw on datasets like the Billion Cell Project, plus work our teams in Chicago and New York are doing to study immune function, with the goal of speeding up the process of testing immune therapies and discovering how to prevent some diseases at the earliest stages.
Expanding our compute cluster
All of this work sits on top of one of the largest compute clusters for nonprofit life sciences research. We’re really proud to contribute this resource to the field of biology, and we’re going to continue investing in it. The cluster will scale up to 10,000 GPUs over the next few years — and help Biohub scientists run increasingly complex simulations of cells and tissues.
Accelerating our rate of progress
As we build more AI models, we also need to create tools to evaluate and improve them. This turns out to be a pretty hard technical problem, which is why we’ve partnered with dozens of leaders in science and technology to try to solve it — and just published a suite of benchmarking tools on our virtual cells platform.
Like our models, the benchmarking toolkit is free and open-source. We’re hoping it will help scientists understand what different models can do, how they perform and where the community can make improvements.
Closing the gap between research, discovery and cures
One of the most important things I learned during my residency at UCSF is that basic science is hope. Research that helps us understand the fundamental biology of cells is by far the biggest driver of breakthrough cures and therapies.
The challenge is that it takes about 30 years to go from one end of that research process to the other. For a lot of patients, and people we love, that isn’t nearly fast enough.
The promise of AI-powered biology is that it’s going to accelerate progress — helping scientists test riskier ideas, run faster experiments and make bigger discoveries. I’m excited that Biohub will help lead the way, and I’m glad to have you with us for the next phase of our journey to cure or prevent all disease.
With care,
