Technology
DiffDock
A state-of-the-art generative diffusion model for highly accurate, blind molecular docking.
DiffDock, developed by MIT researchers, is the first generative diffusion model for molecular docking, reframing the task from optimization to generative modeling. It employs a dual-model architecture: a Score model (20M parameters) generates a diverse set of potential ligand poses, and a separate Confidence model (5M parameters) ranks them for high-likelihood selection. This approach significantly outperforms prior methods: the original DiffDock achieved a 38.2% top-1 success rate (RMSD < 2 Å) on the PDBBind blind docking benchmark, nearly doubling the accuracy of previous deep learning methods. The latest version, DiffDock-L, further boosts this accuracy to 43.0% and demonstrates a 2x speed increase.
What builders pair with DiffDock
Projects using both technologies. Select a pairing to see a project.
4 more pairings
Pairing: 3D visualization
Apple Protein Vision and LLMs for Bioinformatics
Pairing: AlphaFold
Apple Protein Vision and LLMs for Bioinformatics
Pairing: Apple Vision Pro
Apple Protein Vision and LLMs for Bioinformatics
Pairing: ESM-1b
Apple Protein Vision and LLMs for Bioinformatics
Pairing: ESM-2
Apple Protein Vision and LLMs for Bioinformatics
Pairing: ProGen
Apple Protein Vision and LLMs for Bioinformatics
Recent Talks & Demos
Showing 1-1 of 1
