What AI can now do·May 8, 2024·Science

AlphaFold 3 predicts protein-drug interactions at least 50% more accurately

Google DeepMind / Isomorphic Labs

at least 50%Accuracy gain on protein-ligand interactions (PoseBusters)

AlphaFold 3 (Nature) predicted the joint structure of proteins with DNA, RNA, ligands and antibodies with at least 50% better accuracy than existing methods on protein-molecule interactions, the first AI to beat physics-based docking on PoseBusters.

How far it has travelled

From a demonstration to real life, in five steps. Empty steps are the point: a demonstration is not a deployment.

Demonstratedshown once, by its developer or a researcher
yes · May 8, 2024 · Nature paper; at least 50% improvement over existing methods for protein interactions with other molecule types, doubled on some categories; no structural input needed. source
Reproducedshown again by someone else
partial · Nov 11, 2024 · DeepMind released AlphaFold 3 code and weights for academic use on Nov 11, 2024; open reimplementations followed (Boltz-1, Chai-1), but their pages could not be fetched for this entry, so no figures are given. source
Accessibleavailable to the public or to customers
yes · Nov 11, 2024 · AlphaFold Server free for non-commercial research from launch; code and weights for academic use from Nov 2024; commercial use remains restricted. source
Reliableworks most of the time on a defined task
partial · May 8, 2024 · Accuracy claims are DeepMind's own benchmark results; AF3-specific production validation is not documented here. source
Real consequenceshas changed something in the world, for good or ill
partial · May 8, 2024 · The AlphaFold family is in routine research use (200M-structure database; AlphaFold 2 cited 20,000+ times, per the AF3 post); outcomes specific to AF3, such as medicines reaching patients, are not documented here. source

What this could change for you

Drug researchers can model how a candidate molecule binds its target in minutes rather than months; effects on medicines reaching patients are not yet documented.

Sources

primary source.

A milestone is a dated, sourced result that moved what AI can do; a stage is upgraded only when a source shows it (a reproduction, a public release, production evidence, a documented consequence). “Unknown” is the honest default.