
PATCHR
Diffusion-based molecular structure inpainting
for Protein, RNA, DNA, and multi-chain complexes
Download Patchr Studio Desktop
Complete Molecular Inpainting Platform
PATCHR uses synchronized rigid template tracking and local refinement denoising to achieve state-of-the-art accuracy while preserving your experimental coordinates exactly as-is.
End-to-End Workflow
From PDB structure to completed model in Patchr Studio
Structure Inpainting
Fill in missing residues and atoms using diffusion-based generative models with template conditioning.
Protein, DNA, RNA & Complexes
Works with proteins, nucleic acids, protein-ligand complexes, and multi-chain assemblies.
All-Atom Resolution
Generates complete all-atom structures, not just backbone traces. 1.78 Å Cα RMSD on benchmarks.
Simulation-Ready Output
Direct export to GROMACS, AMBER, and OpenMM for molecular dynamics simulations.
Dual Backend Support
Choose between Boltz-2 and Protenix backends. ~10s inference for 200-residue proteins.
CLI + Desktop GUI
Full command-line interface and Patchr Studio desktop app with Mol* 3D viewer integration.
Benchmark Results
Evaluated on 940 PDB40 structures with artificially introduced gaps mirroring real PDB missing-region statistics.
Method Comparison
RMSD (Å) on missing residues — lower is better
| Method | Type | Cα RMSD | All-atom |
|---|---|---|---|
| PATCHR (full) | All-atom | 1.78 | 2.54 |
| Boltz-2 (unmodified) | All-atom | 11.19 | 11.93 |
| Boltz-2 + template + steering | All-atom | 3.22 | 3.89 |
| RFdiffusion2 | All-atom | 9.19 | 10.20 |
| RFdiffusion | Backbone | 2.04 | — |
Structural Context Accuracy
Cα RMSD by secondary structure and solvent accessibility
Get PATCHR
Free and open source under the MIT license. Download the desktop app, install via pip, or run on Google Colab.
Quick Start
git clone https://github.com/DeepFoldProtein/patchr.gitcd patchr && pip install -e .patchr template 1TON allpatchr predict examples/inpainting/1ton_AB.yaml --out_dir results