A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push

Summary · 3 min read

Reeves is Head of Science at Biohub and was a cofounder and chief scientist at Evolutionary Scale, where he worked on the ESM protein language model project. At Biohub, he works toward building a predictive model of cellular biology.

Reeves argues that today’s virtual cell models fall short because biology is difficult to observe and existing data miss important causal and contextual information. He says Biohub’s $1.8 billion Virtual Biology Initiative will expand cellular data and measurement technologies to support an information-theoretic model of cells, with a longer-term goal of predicting and reversing disease states. He expects approaches such as self-supervised pretraining and learning across organisms to help, but says biology also requires empirical experiments and laboratory verification. He is uncertain whether language and biological models will eventually share a latent space, and sees continuous monitoring as a potentially useful complement to Biohub’s focus on fundamental biology.

Minute by minute

1. 4:24

   Aim for a world without disease. Reeves says Biohub’s practical goal is a world without cancer, infectious, chronic, or degenerative disease. He argues that reaching it requires understanding disease mechanisms and moving beyond pills toward programmable medicines that can read and write biological programs.

2. 9:16

   Cell models lack data and scale. Reeves says cells are complex systems, from molecular interactions to their roles in tissues and organisms. He sees limits in both modeling approaches and measurement: transformers lack biology’s multiscale structure, and many relevant cellular features cannot yet be observed or measured at scale.

3. 13:31

   Build an information-level cell model. Reeves says architecture matters when data are scarce, as they are for cellular biology. He describes a possible information-theoretic model that learns how cells respond to inputs, predicting a distribution of next states and eventually modeling biological programs across scales.

4. 22:57

   Coordinate a major data effort. Reeves describes the Virtual Biology Initiative’s flagship partnership with the Department of Energy, NIH, Google DeepMind, Isomorphic, and Meta. He says the effort will build contextual cellular maps, linking molecular measurements with imaging across hundreds of billions or more cells.

5. 26:12

   New optics reveal cellular structure. Reeves says Biohub’s laser phase plate uses a standing-wave laser to create phase contrast in electron microscopy. In cryo-electron tomography, he expects the added contrast to help researchers identify molecules and see how they interact inside cells.

6. 29:21

   Connect biology across scales. Reeves says disease can arise from molecular changes, so models must connect cellular and larger-scale biology. Biohub works across scales, including measuring cell responses in culture and imaging zebrafish to track developing cells and immune-cell interactions in living tissue.

7. 37:21

   Learn across organisms. Reeves says model organisms have revealed mechanisms relevant to human disease, but they are imperfect proxies. He argues that models can transfer information across evolution: ESM and AlphaFold drew on proteins across life, while conserved molecular machinery can support predictions about human biology.

8. 42:17

   Pair pretraining with experiments. Reeves expects self-supervised pretraining to help cellular biology, followed by eliciting the parts of a model relevant to a specific problem. He says biology also needs empirical AI: systems that ask questions, run experiments, learn from results, and improve, with verification in the lab.

9. 46:46

   Map health and disease states. Reeves says cellular atlases across health and disease could show what distinguishes diseases and what causes they share. He expects first drafts within one or two years and a fairly comprehensive map within five years, creating a basis for identifying interventions that could move pathological states toward health.

10. 48:56

    Health records cannot replace causal biology. Reeves says health records could help match existing drugs to patients, but warns that models built only on patient histories may find correlations without linking them to underlying biological causes. He says Biohub is focused on the fundamental biology needed for a step change in understanding and treatment.

11. 51:57

    Language may not capture new biology. Reeves says language can support reasoning and carry knowledge from past biological research, but many important biological facts require experiments and are not represented in text. He is unsure whether language and biological models will share a latent space, noting that fusion would require bridging data and that biology includes concepts without names.

12. 56:31

    Monitoring could link biology to health. Reeves says large-scale personal data could be valuable, especially if continuous monitoring measures more molecular parameters over time. He sees Biohub’s role in building measurement technology; personally, he says he barely tracks his health beyond using a watch to monitor sleep.

Published Oct 20, 2026.