← back to domains
§ 04 · domain

Drug Discovery

Drug Discovery (Small Molecules)

Active learning & Bayesian optimization

1 tools
Closed-loop campaigns

combine generation, prediction, and experimental synthesis/assay in iterative cycles. Used by Recursion, Insitro, Iambic, Schrödinger, Isomorphic Labs (DeepMind spinout).

Cofolding & binding prediction

7 tools
AlphaFold 3

predict the bound complex from sequence + ligand SMILES. Headline improvement: no need for a crystal structure of the apo target.

2024
Boltz

weights publicly released.

Chai

predict the bound complex from sequence + ligand SMILES. Headline improvement: no need for a crystal structure of the apo target.

DiffDock

docking via diffusion / equivariant networks; replacing classical docking in many workflows.

Karma-Dock

docking via diffusion / equivariant networks; replacing classical docking in many workflows.

NeuralPLexer

predict the bound complex from sequence + ligand SMILES. Headline improvement: no need for a crystal structure of the apo target.

Uni-Mol

docking via diffusion / equivariant networks; replacing classical docking in many workflows.

Generative chemistry

4 tools
Flow matching
EquiFM, MolFlow

newer, often better than diffusion.

Pocket-aware diffusion
DiffSBDD, Pocket2Mol, TargetDiff, DecompDiff

generate small molecules conditioned on a binding pocket.

Reinforcement learning for property optimization
REINVENT 4, GFlowNets

widely used in industry.

equivariant generative models
EquiFM, MolFlow

newer, often better than diffusion.

Property & ADMET prediction

5 tools
Chemprop

GNN/transformer property predictors.

MolBERT

GNN/transformer property predictors.

MolGPS

GNN/transformer property predictors.

TDC
Therapeutics Data Commons

standardized benchmark suite.

Uni-Mol2

GNN/transformer property predictors.