Why We’re Going All In on Biology and AI

At the Chan Zuckerberg Initiative, our work has always been guided by a bold mission: to help build a better future for everyone. From the start, we cast a wide net to better understand where our resources and expertise could make the greatest impact.

Over the years, we’ve learned a great deal about where we’re uniquely positioned to help. I’m proud of all of it. We took bold bets, worked alongside incredible partners, and learned where we could be most effective — and where others might be better suited to lead. That journey has led us to sharpen our focus on the intersection of biology and AI, where we’re seeing transformative momentum, while still committing meaningful resources to education and to our local community.

For me, this next science-focused chapter is deeply personal. I was the first in my family to go to college. I studied biology and spent a year teaching elementary school science before going to medical school. I worked as a trainee and pediatrician at UCSF and San Francisco General for eight years. It was there that I saw the limits of medicine and science up close, working with children with rare diseases. For those families, expanding the limits of what we know — advancing basic science research — is their only hope for a better life for their child.

Those experiences shaped not just who I am, but what I believe is possible. And, it informed our mission in science — to help cure, prevent, and manage all disease by the end of this century. This work is already producing results. From our Biohubs to our AI-powered models to the researchers using our tools to uncover new insights, we’re seeing real breakthroughs.

That doesn’t mean our other initiatives weren’t worthwhile. But over time, we’ve learned that trying to do everything makes it harder to deliver deep, lasting impact. We had to focus — not because our values or beliefs changed, but because we needed to decide where our resources and efforts could do the most good. I’m still guided by the same values I’ve lived by my entire life, and that I work every day to pass on to my daughters: optimism for a better future, hard work to make that future possible, and a deep commitment to helping others. The changes we’ve made at CZI will allow us to ensure our work does the most good and achieves the greatest possible impact.

Over the past nine years, we’ve invested more than $7 billion to help communities thrive. Mark and I are proud of what we’ve accomplished — and excited about what’s ahead. With sharper focus, we can accelerate discovery and build the tools science needs to move faster, so that the children I worked with as a pediatrician can have that better life.

By working together, we hope to unlock the promise of AI in science even more rapidly. Here’s how we’re doing it.

Learning To Speak Cell

If you’ve spent any time using AI, you probably know that large language models are prediction machines. Show them billions and billions of words, and they’ll figure out how to put sentences together, write a poem, or (my favorite) summarize the last season of “The Amazing Race.”

Using AI to understand biology means teaching models how to speak the language of cells. Every gene in your DNA is like a word. And by expressing those words in different ways, cells do different things inside your body — from destroying viruses to making your heart beat. For years, scientists — including here at CZI and our Biohubs — have been assembling massive databases, called cell atlases, that show which genes “talk” in different types of cells in many different animals, including ourselves.

We recently released our first AI model that can decipher that language. It’s called TranscriptFormer. If you feed it cell atlas data, TranscriptFormer can tell you what kinds of cells you’re looking at and whether they’re healthy or sick — and in cases where they are sick, it can tell you what those cells are doing to defend themselves.

TranscriptFormer can figure all that out even if it’s never seen cellular data from that organism before. The model doesn’t know what a rhesus monkey is. But even without being told that it’s looking at data from a monkey cell atlas, the model can still tell the difference between a monkey’s liver cells and brain cells. This is useful for health research in predicting whether a finding in one type of species, like mice or monkeys, might translate to human cells before doing experiments in the lab.

Researchers can use TranscriptFormer for free to learn more about the cells they’re studying and find new patterns in their gene expression. This model is an important step in our work to build virtual cells, which will help scientists understand and predict the language of biology like never before.

Watching Cells in Motion

We can learn a lot by studying how cells express their genes. But progress in biology also depends on direct observation — using microscopes to watch how cells move and change in real time.

One of the big challenges here is understanding what we’re looking at. Researchers usually use fluorescent dyes to label different parts of the cell. The images look cool, but manually staining cells takes a long time, and the light used to activate the dyes can damage the cells.

A research team at our San Francisco Biohub has been studying whether AI can improve the process, but a major challenge has been that most models don’t work well across different microscopes or cell types. Our team explored how to make “virtual” staining more adaptable, creating a kind of virtual microscope that can track individual cells without using any dyes. The result, which they just published, is Cytoland: a set of AI models that can identify organelles in the cell across many microscopes and cell types — without any dyes at all.

Virtual staining has a lot of advantages, starting with the fact that it’s faster, cheaper, and easier than the manual process. Even more importantly, it doesn’t degrade the cells — which means we can watch them for longer periods of time, see how they divide and differentiate, and map whole networks of cells as they respond to disease.

Deciphering the Immune System

When we started the Biohub model almost 10 years ago, we put in place an Investigator program where creative scientists, engineers and technologists from across nine different universities are able to pursue bold, visionary ideas. Each of their projects aims to accelerate the pace of scientific discovery and help us realize our mission of curing, preventing and managing disease. We recently named nine Investigators from Columbia University, The Rockefeller University and Yale University to decipher the molecular language of immune cells, enabling these cells to be deployed for disease detection, prevention and treatment in a broad spectrum of age-related diseases, such as neurodegenerative disorders and aggressive cancers. I’m always inspired by our Investigators and look forward to seeing what they accomplish when given the freedom to explore something new in science.

With care,

Co-Founder and Co-CEO
Chan Zuckerberg Initiative