DiffSBDD
SE(3)-equivariant 3D conditional diffusion model for structure-based drug design coupled with AutoDock Vina docking and evolutionary optimization.
The loop
- Proposes
An SE(3)-equivariant diffusion model generates 3D ligand atom coordinates and types directly inside a target protein binding pocket.
- Judged by
RDKit chemical validity checks, synthetic accessibility (SA), QED drug-likeness, and AutoDock Vina / QuickVina2 physics docking score each ligand.
- Steers
An evolutionary partial-noising and denoising loop mutates top-scoring ligands and selects candidates with higher binding affinity and QED.
Headline result
Generated pocket-specific 3D ligands with stronger Vina binding scores and drug-likeness than autoregressive 3D baselines on CrossDocked2020.
What this does not prove
Rigid-pocket docking scores do not account for induced-fit protein conformational changes or explicit water bridges.
How it generalizes the harness
- The algorithm is the solver
- A domain authority judges
- Search over a population
- The generator is not an LLM
An evolutionary docking loop drives an SE(3)-equivariant 3D molecular diffusion generator to optimize pocket binding affinity.
Classification
| What it proposes | Molecule |
|---|---|
| Generative model | SE(3)-equivariant 3D pocket-conditioned diffusion network |
| Algorithmic authority | RDKit chemical validator and QuickVina2 / AutoDock Vina physics docking oracle |
| Feedback | Score |
| What gets updated | Population of candidates |
| When it runs | At inference time |
| Validity | Approximate physics (simulation, not lab results) |
| Evidence | Code released |
| Lineage | Original system |