virtual cell (notes)

09-21

virtual cell - a computational model that can predict cellular behavior under untested conditions. particularly, we want to predict the effects of genetic perturbations, at least as a start.

There are two parts of a virtual cell - the foundation and the scaffolding

the actual virtual cell

the actual virtual cell - we want a working model. current models do not outperform simple linear baselines (Ahlmann-Eltze et. al). one possible bottleneck is that we are short on data - specifically, causal data on genetic perturbations. quality data is difficult to collect; different labs in different locations can report different results for the same perturbation.

scaffolding ("agentic workflows")

suppose we have a functional virtual cell. we want to build the tools to help us use it effectively. we've had more success in building scaffolding for adjacent problems - see the Virtual Lab (Swanson et. al) and Paper2Agent (Miao et. al).

long-term vision

a virtual cell meaningfully and accurately predicts the behavior of a real cell. we use AI agents to probe this cell since there is too much information for any single scientist to hold or query.