AI offers a huge amount of leverage for biology, yet there's a significant lack of foundational tools in the space. It is surprising that in 2025, there is no equivalent to the periodic table of elements for biology. Building these types of tools is one of the most important things that can be done.

When the goal to cure and prevent disease by the end of the century was first announced, many scientists couldn't take it seriously. They thought it was crazy because the existing path, which involved funding the next best grant for every lab, would never lead to that outcome. This reaction highlighted a disconnect between the biology and AI communities.

The biology folks looked at it as if it were crazy ambitious. And then the AI folks are like, well, that's kind of boring. That's just automatically going to happen. There's something in between there that needs to be bridged.

To make progress, the scientific community needs fundamentally new tools to cure disease, not just more funding.

Priscilla's motivation to improve people's lives through medicine was shaped by her time as a pediatrician. She often encountered young patients and families for whom there were no clear answers or cures. It was her job to translate a few lines of information from a printout into a care plan. This experience highlighted the critical need for basic science to push the boundaries of what is medically possible, which she describes as the "pipeline of hope."

This led to the ambitious goal of helping to cure all diseases. Mark clarifies that the strategy is not for them to cure the diseases themselves, but to empower the scientific community by accelerating the pace of basic science. The core theory is that major scientific breakthroughs are almost always preceded by the invention of new tools that allow for new ways of observing phenomena, like the microscope or telescope. Their approach is to build the tools that will help the entire field advance.

This philanthropic strategy fills a specific niche. Most scientific funding from government grants is parceled into smaller amounts for near-term projects. CZI focuses on longer-term, more expensive tool development, which can cost hundreds of millions to a billion dollars over 10 to 15 years. These tools are then given to the scientific community.

Initially, many scientists thought the goal to cure all disease was crazy. But, as Priscilla explains, when they were asked why it was impossible, they identified the real barriers: a lack of shared tools and large-scale data sets. This transformed a seemingly impossible goal into a concrete problem to solve. Mark adds that there was a funny split in perception: biologists saw it as wildly ambitious, while AI experts thought it was an inevitable and somewhat boring outcome.

The work today lives at the intersection of "frontier biology paired with frontier AI." The goal is to produce specific data sets for the express purpose of training AI models to build things like virtual cells. This is a unique approach, as even groundbreaking models like AlphaFold were built on public data sets created decades ago. This focus on science has yielded the biggest returns for their philanthropic efforts, leading them to double down and make their Biohub the primary focus.

When selecting grand challenges with a 10 to 15-year time horizon, the goal is to find projects with a credible pathway, but not one where everything is already solved. There needs to be enough ambiguity to allow for risk and the potential for higher-than-expected returns. This approach requires a strong leader at the helm who can navigate that uncertainty.

This strategy is put into practice through three biohubs, each with a specific focus. The New York hub works on cell engineering, exploring how to engineer cells to detect information or take specific actions. In Chicago, the focus is on building tissues and understanding cell communications within them. The San Francisco hub concentrates on deep imaging and transcriptomics.

The locations are chosen strategically based on partner universities. Researchers from these academic institutions collaborate at the biohubs in an interdisciplinary environment that is less constrained than a traditional lab. This model leverages the strengths of both the dedicated biohubs and the supporting academic labs.

The recent emergence of large language models has been a significant development. While tools were already in place to measure and collect interesting data, the means to fully analyze these large datasets were not yet clear. Large language models provided the breakthrough needed to make sense of all the collected information.

Success for the CZ Bio Hub is envisioned as fostering an explosion of a community building the next wave of precision medicine for both rare and common diseases. Priscilla Chan argues that most diseases should be treated as rare diseases because each person's biology is different. Currently, treatments for common conditions like hypertension or depression are often a process of trial and error. The goal is to move towards precisely, accurately, and quickly treating people by understanding their individual biology.

This is achieved by connecting a genetic mutation to its downstream cellular effects and protein expression. This allows researchers to identify specific drug targets and predict potential side effects by understanding where else a drug might interact with the body. The Bio Hub focuses on enabling this basic science, creating models that others can use to build diagnostics and therapeutics.

Vineeta Agarwala shares an example of this in action. A startup working on idiopathic pulmonary fibrosis, a disease with an unknown cause, used the Bio Hub's open-source cell by gene atlases. By analyzing millions of single cells from patients, they were able to pinpoint specific fibroblasts and their gene expression. This data helped them identify potential new drug targets for a disease that was previously a mystery. This demonstrates the power of making complex biological data accessible through user-friendly software and tools, which is a core part of the Bio Hub's approach.

The cell atlas project was inspired by the absence of a biological equivalent to the periodic table of elements. Mark explains the goal was to create a standardized format for biological data. While the initial focus was on data standardization, the idea of using this data to build virtual cell models emerged later as AI technology advanced.

I mean, the cell atlas work overall and it's kind of this crazy thing that we're here in 2025 and there's not the kind of periodic table of elements equivalent for biology.

Priscilla shares the origin story of Cell by Gene, which began 10 years ago. The project started by funding the methodology for single-cell analysis to ensure standardization. However, a bottleneck quickly appeared: researchers could not annotate the data fast enough. To solve this, the team built Cell by Gene as an annotation tool. Because everyone began using the same tool, their data was automatically standardized.

This shared tool created a community and a network effect. Scientists started sharing their data back, contributing to a massive, shared atlas. This community-driven effort was incredibly successful.

We only funded 25% of it. 75% came from the broader community saying this is useful and there's an easy way for us to standardize and build these together.

This dynamic was likened to the internet's growth. The value proposition was clear: come for the annotation tool, and stay for the comprehensive virtual cell model and shared data. The initial decision to enforce a consistent format was crucial, as it made all the contributed data portable and usable, driving widespread adoption.

One of the most important tools being developed is the virtual cell. The goal is to build up a hierarchy of models, from proteins to cell structures to entire systems like a virtual immune system. These models will help scientists generate hypotheses for their work before running full experiments in a lab. Mark explains this will be useful for precision medicine and other scientific research.

Priscilla adds that virtual cell models allow scientists to pursue riskier ideas. Traditional wet lab work is expensive and slow, which naturally discourages high-risk, high-reward projects. A researcher needs to ensure a certain success rate to maintain their career.

But if you had a virtual cell model where you could simulate really high quality biology, you could actually then start testing at and tinkering on the computational side and ask riskier questions, things that would have been expensive and costly in terms of time and resources to do in the lab.

This allows researchers to see if an idea has promise in silico before making the significant time and money investment in a wet lab. The virtual cell acts as a new kind of model organism, like "the new fruit fly," but with direct fidelity to the human body. It doesn't need to be perfectly accurate to be useful. As the classic saying goes, "All models are wrong. Some are useful." Even partial de-risking makes the process more efficient.

The process of creating these models is similar to developing language models. Different Biohubs focus on specific challenges, like cellular engineering or inflammation, and build tools to generate novel datasets. These datasets are then used to build models, which are eventually combined into an increasingly general virtual cell. The approach is hierarchical. The view is that you need to build a state-of-the-art protein model and then integrate that into a state-of-the-art cellular model. This provides a deeper, hierarchical understanding of how the sub-components of cells interact, which is essential for modeling complex systems and personalization.

Mark and Priscilla announced a significant strategic shift for the Chan Zuckerberg Initiative (CZI). Previously decentralized, their Biohubs, software development, and AI research are now coming together as a single, unified team called the Biohub. This operating philanthropy will focus on advancing biology at the intersection of AI.

Mark explained that CZI is doubling down on science, as it's where they feel they've made the biggest difference. The Biohub will be the main thrust of their philanthropy going forward. He believes that with advances in AI, their initial goal of helping the scientific community cure and prevent all diseases by the end of the century can be achieved significantly sooner.

Priscilla elaborated on the rationale for this unification. While many groups excel at either frontier AI or frontier biology, CZI aims to uniquely tie the two together. They are building instrumentation and generating massive datasets, such as mapping the cell at a nearly atomic level. By integrating their AI model builders with their data generators, they can create a powerful flywheel. If a model shows gaps, the team can immediately work on creating the next dataset to fill them. This requires researchers to work in close collaboration.

It's more than just writing down a spec and saying, 'Please deliver this.' These people need to be sort of working shoulder to shoulder and shaping each other's work for this to actually be the more and more accurate model of how the human cell works.

Ben noted this aligns with a counterintuitive trend in the wider AI industry: domain-specific models are proving incredibly effective, often more so than general models. Knowing the specific problem you're solving is critical. Priscilla agreed, emphasizing that the conversation with the scientist and the user interface are also vital. For example, CZI's Cell by Gene tool was intentionally designed with a low barrier to entry to encourage interdisciplinary collaboration. They hope future models will further lower this barrier, allowing experts from diverse fields, like immunology and neurodegeneration, to contribute to each other's work.

When approaching philanthropy in science, the goal is to be as additive as possible by identifying what is underrepresented. Science is typically very decentralized. In response, a lot of value can be unlocked by encouraging collaboration in ways that may seem simple but were not happening before. The first Biohub was a collaboration between UCSF, Stanford, and Berkeley, creating a formal structure for smart people from different institutions to work together.

Another key aspect was fostering cross-disciplinary work by having biologists sit next to engineers. Mark notes the power of physical proximity in solving organizational issues.

It's interesting how many organizational questions or problems you can fix just by having two teams sit together. It's like it doesn't matter what the org chart is or whatever. It's like you guys need to sit next to each other and until you get this thing to work.

This centralized, collaborative model isn't meant to be how all science should work, but it fills a space that wasn't the default. Much of the inspiration came from physics, where labs historically rallied around big projects and shared resources.

Priscilla explains that the modern expansion of a lab is not necessarily about square footage, but about compute power. Researchers increasingly just want access to GPUs. The Biohub was among the first to build a large-scale compute cluster, moving from thousands to a planned 10,000 GPUs. This resource allows for different types of questions to be asked. The Biohub also invites outside scientists to apply and use this resource, seeding new collaborations.

When reflecting on the past 10 years, Priscilla Chan noted she was initially envious of for-profit companies because of their clear market feedback. It is much harder to determine if you are doing a good job in philanthropy without those signals. However, a key insight came after a decade of work: their projects not only achieved their stated goals but delivered even more value than anticipated. This became the clear signal to double down on their work in biology.

A guiding principle for their future is to tolerate early ambiguity. This involves being patient with a long time horizon while also being impatient with the short-term iterations along the way. Priscilla explains that this constant work is what allows an organization to get lucky. For instance, their previous efforts in building datasets perfectly positioned them to take advantage of recent advancements in AI and large language models.

This is a space that there's just gonna be a huge amount of leverage with AI. And it still seems like there could be a lot more effort in this space around building tools and just accelerate the whole thing a lot better.

Mark Zuckerberg expanded on this, noting that while they don't have traditional customers, they get feedback by seeing how many scientists use the tools they build. Their work fills a unique void in the scientific pipeline. This pipeline includes accelerating basic science, biotech development, and large-scale pharmaceutical distribution. Their specific role is creating foundational tools that accelerate the entire process, a space with immense leverage that others are not filling. This unique contribution answers a key question for them: "If we didn't exist, would it be a problem?" The answer is a clear yes.