MultiGeo-DTA

• MultiGeo-DTA
  • Overview
  • Installation Guide
  • Dependencies
  • Data availability
  • Test
  • Train from scratch
  • Other usages
  • Contact

Overview

MultiGeo-DTA is a multimodal neural network that integrates structure and sequence information to predict compound–protein binding affinity.

Installation Guide

git clone https://github.com/liyazi712/MultiGeo-DTA.git
cd MultiGeo-DTA

Dependencies

This package is tested with Python 3.8 and CUDA 11.8 on Ubuntu 20.04, with access to an Nvidia V100 GPU (32GB RAM), AMD EPYC 7443 CPU (2.85 GHz), and 512G RAM. Run the following to create a conda environment and install the required Python packages (modify pytorch-cuda=11.8 according to your CUDA version).

conda create -n MultiGeo-DTA python=3.8
conda activate MultiGeo-DTA

# (pip or conda, select one)
conda install pytorch==2.1.0 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install torch==2.1.0 --index-url https://download.pytorch.org/whl/cu118

conda install -c dglteam/label/th21_cu118 dgl

pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.1.0+cu118.html
pip install dgl -f https://data.dgl.ai/wheels/torch-2.1/cu118/repo.html
pip install rdkit pyyaml scikit-learn torch_geometric pandas joblib
pip install causal_conv1d-1.4.0
pip install mamba_ssm-2.2.2

Install causal_conv1d and mamba_ssm may has error, then you can download the .whl files and install as following：

pip install causal_conv1d-1.4.0+cu118torch2.1cxx11abiFALSE-cp38-cp38-linux_x86_64.whl
pip install mamba_ssm-2.2.2+cu118torch2.1cxx11abiFALSE-cp38-cp38-linux_x86_64.whl

The example provides versions of causal_conv1d and mamba_ssm compatible with PyTorch 2.1.0 (CUDA 11.8) and Python 3.8. If your setup differs, ensure PyTorch, PyTorch Geometric, and DGL versions are aligned. For other versions, download causal_conv1d and mamba_ssm from their respective GitHub releases:

• mamba_ssm: https://github.com/state-spaces/mamba/releases
• causal_conv1d: https://github.com/Dao-AILab/causal-conv1d/releases

common error:

1. OSError: libcusparse.so.11: cannot open shared object file: No such file or directory
2. solution: conda install -c nvidia cudatoolkit=11.8

Running the above lines of conda install should be sufficient to install all MultiGeo-DTA's required packages (and their dependencies).

Data availability

1. Download open source data from Hugging Face Dataset. (Because of the official website's limitation, PDBBind v2021 dataset will open source after the official website open source them)

   pip install -U huggingface_hub
   export HF_ENDPOINT="https://hf-mirror.com"
   cd create_dataset
   huggingface-cli download laddymo/MultiGeoDTA --repo-type dataset --local-dir MultiGeoDTA --local-dir-use-symlinks False

Test

1. PDBBind_v2016

python test_MultiGeoDTA.py --task pdbbind_v2016 --output_dir ./MultiGeoDTA/output/pdbbind_v2016 --model_file pdbbind_v2016

2. PDBBind_v2020

python test_MultiGeoDTA.py --task pdbbind_v2020 --output_dir ./MultiGeoDTA/output/pdbbind_v2020 --model_file pdbbind_v2020

3. PDBBind_v2021_time

python test_MultiGeoDTA.py --task pdbbind_v2021_time --output_dir ./MultiGeoDTA/output/pdbbind_v2021_time --model_file pdbbind_v2021_time

4. PDBBind_v2021_similarity

python test_MultiGeoDTA.py --task pdbbind_v2021_similarity --output_dir ./MultiGeoDTA/output/pdbbind_v2021_similarity/new_new/0.5 --model_file pdbbind_v2021_similarity/new_new/0.5 --split_method new_new --thre 0.5 

5. LP-PDBBind

python test_MultiGeoDTA.py --task lp_pdbbind --output_dir ./MultiGeoDTA/output/lp_pdbbind --model_file lp_pdbbind

5. ZINC(virtual screening)

python run_vs.py --output_dir ./MultiGeoDTA/output/zinc --model_file pdbbind_v2020 --device 0

note: split_method: new_new, new_compound, new_protein; thre: 0.3, 0.4, 0.5, 0.6, modify output_dir and model_file according to split_method and thre)

Train from scratch

1. PDBBind_v2016

python run_MultiGeoDTA.py --task pdbbind_v2016 --output_dir ./MultiGeoDTA/output/pdbbind_v2016

2. PDBBind_v2020

python run_MultiGeoDTA.py --task pdbbind_v2020 --output_dir ./MultiGeoDTA/output/pdbbind_v2020

3. PDBBind_v2021_time

python run_MultiGeoDTA.py --task pdbbind_v2021_time --output_dir ./MultiGeoDTA/output/pdbbind_v2021_time

4. PDBBind_v2021_similarity

python run_MultiGeoDTA.py --task pdbbind_v2021_similarity --output_dir ./MultiGeoDTA/output/pdbbind_v2021_similarity/new_new/0.5 --split_method new_new --thre 0.5

note: split_method: new_new, new_compound, new_protein; thre: 0.3, 0.4, 0.5, 0.6, modify output_dir and model_file according to split_method and thre)

5. LP-PDBBind

python run_MultiGeoDTA.py --task lp_pdbbind --output_dir ./MultiGeoDTA/output/lp_pdbbind

Other usages

1. missing_dataset. train:

   python run_MultiGeoDTA.py --task pdbbind_v2016 --output_dir ./MultiGeoDTA/output/pdbbind_v2016_robustness/missing_0.2
   

   test:

   python test_MultiGeoDTA.py --task pdbbind_v2016 --output_dir ./MultiGeoDTA/output/pdbbind_v2016_robustness/missing_0.2 --model_file pdbbind_v2016_robustness/missing_0.2
   

2. noise_label. train:

   python run_MultiGeoDTA.py --task pdbbind_v2016 --output_dir ./MultiGeoDTA/output/pdbbind_v2016_robustness/noise_0.2
   

   test:

   python test_MultiGeoDTA.py --task pdbbind_v2016 --output_dir ./MultiGeoDTA/output/pdbbind_v2016_robustness/noise_0.2 --model_file pdbbind_v2016_robustness/noise_0.2
   

Contact

Please submit GitHub issues or contact Yazi Li ([email protected]) for any questions related to the source code.