Machine Learning Molecules Systems Biology

Machine Learning
‍Systems Biology

Scroll Down
Long-Term Plan


Our goal is to simulate in a bottom-up fashion, by 2050, a simple metazoan cell with sufficient fidelity such that any experimentally measurable cell-biological quantity can be predicted in silico with comparable accuracy, if at possibly greater cost. Such simulations would advance synthetic biology and have therapeutic potential; their pursuit may reveal new science. What such a simulation will look like is perhaps the defining question of our research program. Direct physical simulation at the atomic level is unlikely to ever be scalable to whole cells (quantum computers notwithstanding) and even if it were, is unlikely to yield human-interpretable insights. Our task then is to find the right abstractions, conceptual and mathematical, to meaningfully simulate and understand biology. In many ways this is the most interesting part of our research program; discovering life’s computational paradigms and primitives—the simulation part is just an excuse to do it. Having a concrete long-term goal does however serve an important purpose by acting as an organizing fulcrum, prioritizing our research pursuits and motivating us with a long-term vision. It is our North Star.

More important than the destination is our journey, and in particular what we expect to build and learn along the way. On the building side we construct, for now, machine learning models of biomolecules and their interactions, and compositions of such models to derive representations of more complex biological phenomena. On the learning side, we use these models to understand the organization and logic of (for now) metazoan signaling networks, how they vary in human populations, and how they are dysregulated in cancer. While our long-term goals are basic, maintaining translational anchors ensures that our journey is productive along the way.