RMSD-Pred

Protein-Ligand Binding Pose RMSD Prediction using Graph Neural Networks

Overview

RMSD-Pred scores the quality of a docked protein-ligand pose by predicting its RMSD to the (unknown) native binding mode — without needing the crystal pose. It builds three graphs per complex — protein pocket, ligand, and the protein-ligand interface — and couples them at every layer.

How it works

1. Pocket extraction — protein atoms with a ligand atom within 10 Å.
2. Graph construction — protein graph (atom pairs within 4 Å, distance RBF edges; nodes encoded by residue + atom-type embeddings), ligand atom graph (covalent bonds), and a complex interface graph (atom pairs within 5 Å, distance RBF). Each node carries a 20-step random-walk positional encoding (LSPE).
3. Message passing — 3 parallel stacks (protein / ligand / complex) of 4 GatedGCNLSPE layers (emb=256). Protein and ligand states are stitched into the complex graph and fed back each layer.
4. Readout — sum-pool complex nodes → MLP. Two heads share the architecture: a regression model (predicted RMSD in Å) and a classification model (probability that RMSD > 2 Å).

Companion model: BA-Pred predicts binding affinity (pKd). Typical pipeline: dock → filter poses with RMSD-Pred → rank affinity with BA-Pred.

Installation

git clone https://github.com/eightmm/RMSD-Pred.git
cd RMSD-Pred
uv sync   # installs PyTorch (CUDA 12.8) + PyTorch Geometric

GPU (incl. NVIDIA Blackwell / sm_120): uv sync uses the bundled CUDA 12.8 index. With pip: pip install -e . then pip install torch --index-url https://download.pytorch.org/whl/cu128

Usage

Command Line

rmsdpred -r example/prot.pdb -l example/ligs.sdf -o results.tsv --device cuda

Python API

from rmsdpred.inference import inference

# Uses packaged seed0 reg + cls weights by default
inference(
    protein_pdb="example/prot.pdb",
    ligand_file="example/ligs.sdf",
    output="results.tsv",
    batch_size=128,
    device="cuda",
)

# Or specify custom weights
inference(
    protein_pdb="example/prot.pdb",
    ligand_file="example/ligs.sdf",
    output="results.tsv",
    batch_size=128,
    reg_weight="/path/to/reg.pth",
    cls_weight="/path/to/cls.pth",
    device="cuda",
)

Parameters

Parameter	Description	Default
-r, --protein_pdb	Receptor protein PDB file	required
-l, --ligand_file	Ligand file (.sdf / .mol2 / .dlg / .pdbqt / .txt)	required
-o, --output	Output TSV file	result.tsv
--reg_weight	Regression model weight file	packaged random_reg_seed0.pth
--cls_weight	Classification model weight file	packaged random_cls_seed0.pth
--batch_size	Batch size	128
--ncpu	CPU threads / DataLoader workers	4
--device	cuda or cpu	cuda

Input/Output Formats

Input

• Protein: PDB format (.pdb)
• Ligands: one of
  • .sdf — SD file (multi-molecule / multi-pose supported)
  • .mol2 — Tripos MOL2 (multi-molecule supported)
  • .dlg / .pdbqt — AutoDock poses (parsed via Meeko); ADG_Score is taken from the docking log when available
  • .txt — a list of any of the above file paths, one per line

Output

Tab-separated file with columns:

• Name — pose identifier (_Name property if present, else <file>_<index>)
• pRMSD — predicted RMSD to native pose (Å)
• Is_Above_2A — probability that RMSD > 2 Å (0–1)
• ADG_Score — AutoDock score (NaN if unavailable)

Poses that fail to parse/build yield NaN.

Weights

Packaged checkpoints live in src/rmsdpred/weight/random/: regression (random_reg_seed{0,1,2}.pth) and classification (random_cls_seed{0,1,2}.pth). Defaults are the seed0 pair. Average across seeds for ensemble predictions.

Project Structure

RMSD-Pred/
├── src/
│   └── rmsdpred/          # Main package
│       ├── data/          # Graph construction, atom/residue features, loaders
│       ├── model/         # GatedGCNLSPE, PredictionRMSD (PyG)
│       ├── weight/        # Packaged reg + cls weights
│       └── inference.py   # Inference engine + CLI
├── example/               # Sample protein + ligand poses
├── tests/                 # Smoke tests (dataset build + forward)
├── pyproject.toml
└── README.md

License

This project is licensed under the Apache License 2.0.

Citation

@article{Sim_2026,
  title={BA-Pred and RMSD-Pred: Integrated Graph Neural Network Models for Accurate Protein-Ligand Binding Affinity and Binding Pose Prediction},
  author={Sim, Jaemin and Lee, Juyong},
  journal={Journal of Chemical Information and Modeling},
  year={2026},
  month={apr},
  doi={10.1021/acs.jcim.5c02591},
  url={https://doi.org/10.1021/acs.jcim.5c02591}
}