Mapping the four data gaps between AI drug-discovery models and wet labs

QuanquanGu · x · 2026-11-14

A founder-oriented thread mapping what AI drug-discovery models need from wet labs versus what labs actually deliver.

Four model groups, four evidence types: understanding cell state needs observations with source and conditions; predicting outcomes needs endpoints and controls that match the prediction; design needs each design linked to whether it was made and whether it worked; experiment selection needs the record of how each choice was made.
The bottleneck isn't measurement: labs can already measure most of this. The real gaps fall into four buckets — missing measurement, delivery, linkage, and rights — each a different product with a different price.
For founders: count the unit your model must generalize to. Teams are already commissioning data to their own spec, like Biohub's Billion Cells Project, Recursion's in-house Cell Painting, and Arc Institute's State perturbation model.

More from Venture