DSMBind

This is the official Github repo for the following papers:

• Jin et al., DSMBind: SE(3) denoising score matching for unsupervised binding energy prediction and nanobody design, Biorxiv 2023
• Jin et al., Unsupervised protein-ligand binding energy prediction with Neural Euler's Rotation Equations, NeurIPS 2023

Data

The training and test data can be downloaded from https://zenodo.org/records/10402853. Please put the download data in data/ under the root folder.

Installation

Please make sure you install the following packages

• pytorch (tested on v1.13)
• biotite (https://www.biotite-python.org/install.html)
• SRU++ (https://github.com/asappresearch/sru)
• ESM-2 (https://github.com/facebookresearch/esm)
• Chemprop (https://github.com/chemprop/chemprop)
• tqdm
• rdkit
• sidechainnet <= 0.7.6 (1.0 changes the side-chain dimension from 14 to 15)

Once you finished these installation, please run pip install -e .. To install SRU++, run

git clone https://github.com/asappresearch/sru
cd sru
git checkout 3.0.0-dev
pip install .

Inference

This notebook tutorials.ipynb contains all the relevation information on predicting binding energy for protein-ligand, protein-protein, and antibody-antigen complexes.