Our mission is to build world models for biology that can truly represent the complex dynamics of nature.
Today's models represent biology as a distribution over the data they were fed, and reasoning inside them is sampling from that distribution. It is treated the same way as a language model. But biology is way more than this. There is a mechanism underneath the data and a deeper structure that next token prediction alone does not capture.
The issue is that we do not yet know what it means for two molecules to interact, or at what level that interaction should be described. Quantum, molecular, or mechanical. Proteins, RNA, genomes, or atoms. There's no unified theory that can explain the interactions and mechanisms of biology. This is especially challenging because biological data is inherently scarce, noisy, and incomplete. We observe only a tiny fraction of the possible states of a biological system, and for much of the data we collect, we still do not know what it means. So current models, which approximate function from a distribution, collapse the moment something unseen arises. Biology deserves its own research, its own theorems, algorithms, and architectures.
The natural world is governed by physics, and those data already contains those laws within it. If we can build models that learn the true functions governing how biological systems move, interact, and evolve, we can move beyond approximating the system toward actually finding it. Physics-grounded models could allow us to reconstruct the interactions within cells with extraordinary fidelity, ultimately enabling truly functional virtual cells and, beyond them, increasingly complete models of living systems. Biology remains one of the most complex and least understood frontiers of science. Learning to simulate it from first principles would not only deepen our understanding of nature, but transform our ability to predict disease, design interventions, and advance human health.