EquiDock addresses the computationally intensive problem of rigid 3D protein-protein docking using geometric deep learning. It employs independent SE(3)-equivariant models to achieve fast and accurate predictions, targeting researchers and engineers in computational biology and drug discovery. The primary benefit is accelerating the docking process, crucial for understanding molecular interactions.

How It Works

The core of EquiDock lies in its use of SE(3)-equivariant neural networks, which respect the rotational and translational symmetries of 3D space. This geometric deep learning approach processes protein structures, represented as graphs of residues, to predict binding poses. The system is designed for end-to-end rigid docking, handling data preprocessing, model training, and inference.

Quick Start & Requirements

• Installation: Primarily involves cloning the repository and setting up Python dependencies.
• Prerequisites: Python 3.9.10, CUDA 10.1, PyTorch 1.10.2, DGL 0.7.0, RDKit 2021.09.4.
• Operating System: Linux/macOS only; Windows requires modifications.
• Data: Requires downloading and preprocessing the DB5.5 and DIPS datasets using provided scripts (preprocess_raw_data.py).
• Training: CUDA_VISIBLE_DEVICES=0 python -m src.train -hyper_search
• Inference: python -m src.inference_rigid
• Links: DB5.5 dataset: https://zlab.umassmed.edu/benchmark/; DIPS dataset: https://github.com/drorlab/DIPS, https://github.com/amorehead/DIPS-Plus. DIPS data download: https://www.dropbox.com/s/sqknqofy58nlosh/DIPS.zip?dl=0.

Highlighted Details

• Utilizes novel SE(3)-equivariant models for protein docking.
• Offers an end-to-end solution for rigid docking tasks.
• Provides pretrained models and scripts to reproduce paper results.

Maintenance & Community

The project is associated with academic research (ICLR 2022) but lacks explicit details on ongoing maintenance, community channels (e.g., Discord/Slack), or a public roadmap.

Licensing & Compatibility

The provided README does not specify a software license. This absence creates ambiguity regarding usage rights, commercial compatibility, and derivative works.

Limitations & Caveats

The codebase is explicitly stated to work only on Linux and macOS, requiring modifications for Windows compatibility. Potential steric clashes are noted as an issue, addressed via a postprocessing step, indicating that raw model outputs may not always be physically realistic without this intervention. A specific CUDA version (10.1) is required, which might pose compatibility challenges with newer hardware or software stacks.