Research map / Probabilistic and scientific modeling / Molecules and proteins
Taming Generative Co-Folding Prior for Molecular Docking with Diffusion Bridge
Shikun Feng, Zipeng Yang, Qiuyi Li, Mingze Yin, Xiaoyuan Zhang, Tingjun Hou, Yiheng Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes adapting a generative co-folding prior (AlphaFold 3-style) to molecular docking via a diffusion bridge, based only on the truncated abstract.
Problem
Molecular docking predicts protein-ligand complex 3D structure conditioned on an input protein (apo for flexible, holo for rigid docking); co-folding models like AlphaFold 3 are accurate but the abstract is truncated before stating their limitation.
Approach
Uses a generative co-folding prior and a diffusion bridge to connect the input receptor state to the predicted complex; specific mechanism not stated in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| model docking as apo-to-holo diffusion bridge instead of noise-to-structure | Protenix-Mini AF3 baseline · PoseBusters 20 steps | lDDT (%): 84.79 -> 90.41 | Improved |
| single-step BridgeDock vs 20-step AF3 baseline | BridgeDock · PoseBusters 1 step | % ligand RMSD < 2A: 68.61 (20-step AF3) -> 69.86 (1-step) | Improved |
| apo-conditioned fine-tuning without bridge (AF3-SFT-Apo) (ablation) | AF3 baseline · PoseBusters 20 steps | % ligand RMSD < 2A: 68.61 -> 70.14 | Improved |
| add apo-to-holo bridge on top of apo conditioning (ablation) | AF3-SFT-Apo · PoseBusters 20 steps | % ligand RMSD < 2A: 70.14 -> 73.21 | Improved |
| add AF3 alignment (reparameterize bridge states to AF3 denoising schedule) (ablation) | BridgeDock · PoseBusters 20 steps | % ligand RMSD < 2A: 68.42 -> 73.21 | Improved |
| add AF3 alignment (ablation) | BridgeDock · PoseBusters 20 steps | mean ligand RMSD (A): 2.77 -> 2.26 | Improved |
| BridgeDock vs Protenix baseline on median RMSD | BridgeDock · PDBBind test | median ligand RMSD: 0.99 -> 1.09 | Hurt |
| vary diffusion-time offset delta from 1 to 0.001 (ablation) | BridgeDock · PoseBusters 20 steps | % ligand RMSD < 2A: 71.77 to 73.21 | No clear effect |
| use BridgeDock with larger Protenix-Base v1.0.0 backbone | BridgeDock-v1 · FoldBench protein-ligand | success rate (%): 56.97 -> 59.33 | Improved |
| BridgeDock vs pocket-aware docking baselines | BridgeDock · PDBBind test | % ligand RMSD < 2A: 42.90 (Re-Dock) -> 71.00 | Improved |
Limitations
- Full text unavailable; only truncated abstract was available
Experimental setup
Builds on: AlphaFold 3
Closest papers by meaning
- Re-Dock: Towards Flexible and Realistic Molecular Docking with Diffusion Bridge (ICML 2024 · similarity 0.91)
- Bridge-IF: Learning Inverse Protein Folding with Markov Bridges (NeurIPS 2024 · similarity 0.87)
- Few-Step Cofolding with All-Atom Flow Maps (NeurIPS 2026 · similarity 0.87)
- DeltaDock: A Unified Framework for Accurate, Efficient, and Physically Reliable Molecular Docking (NeurIPS 2024 · similarity 0.87)
- Deep Confident Steps to New Pockets: Strategies for Docking Generalization (ICLR 2024 · similarity 0.86)
- DiSCO: Diffusion Schrödinger Bridge for Molecular Conformer Optimization (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
