Dynamic Proteome

The thesis that static structural snapshots of proteins are necessary but insufficient for drug discovery — the next leap requires modeling proteins as dynamic machines operating on millisecond timescales. Championed by Karen Akinsanya (Schrödinger) at Session I.

Why Static Structures Hit a Ceiling

• Crystallography and cryo-EM have made rational design genuinely rational — but proteins are machines working on millisecond time
• Cryptic pockets open and close; side chains rotate; none of this is sampled by a crystal
• The SARM1 case study: a previously invisible pocket appeared only when a designed molecule engaged, and dose-dependent dynamics created a clinical liability no classical gate caught

The Data Gap

• The training data for “dynamic biology” mostly doesn’t exist yet
• Whoever builds those datasets — with closed-loop autonomous experiments — owns the next decade
• Public chemistry and structure data are saturated; the next wave comes from generating new data

Three Blind Spots

1. Edges — ~92% of the PPI network unmapped; only ~4% of ~650,000 PPIs have structures
2. Atoms — biochemistry runs at ~300× dilution vs. the crowded cytoplasm
3. Drugs — median PFS gain across 234 cancer drug approvals (2003–2021) is ~3.3 months

Cross-References

• AI in drug discovery — algorithms are ahead of the data
• Productivity paradox — dynamics is part of the explanation for why more science hasn’t produced more drugs