ChemGLaM: Chemical Genomics Language Models for Compound-Protein Interaction Prediction

---

Table of Contents

• ChemGLaM: Chemical Genomics Language Models for Compound-Protein Interaction Prediction
  • Description
  • Directory Structure
  • Environment Setup
  • Usage
    • Finetuning for CPI datasets
    • Finetuning for classification with Evidential Deep Learning (EDL)
    • Inference with the finetuned model
    • Run with Docker
  • Contact
  • Reference

---

Description

ChemGLaM is a large language model for Compound-Protein Interaction Prediction.

Directory Structure

.
├── cache                : Directory for protein embeddings
├── chamglam             : Source code for ChemGLaM
├── config               : Config examples
├── data                 : Directory for datasets 
├── figures              : Figures for README.md
├── logs                 : Output Directory
├── script               : Script examples
├── Dockerfile           : Docker file
├── LICENSE              : License file
├── predict.py           : main script for prediction
├── prediction_demo.ipynb: example for inference one pair
├── README.md            : This file
├── setup.py             : setup file
└── train.py             : main script for fine-tuning

Environment Setup

To set up the environment for this project, follow these steps:

1. Create a new conda environment named chemglam with Python 3.11:

   conda create -n chemglam -y python=3.11

2. Activate the environment:

   conda activate chemglam

3. Install the required dependencies:

   pip install -e .

4. Install RDKit (version 2024.9.2) from the conda-forge channel:

   conda install -c conda-forge rdkit=2024.9.2

Usage

Finetuning for CPI datasets

The following is the example of the command for finetuning ChemGLaM with CPI datasets.

python train.py -c /path/to/config.json

Finetuning for classification with Evidential Deep Learning (EDL)

By setting the argument "evidential" as true in a config file, you can run EDL for classification tasks.

python train.py -c /path/to/edl_config.json

Inference with the finetuned model

Coming Soon.
You can use the inference script by specifying the --checkpoint_path with the path to the finetuned model and setting deterministic_eval: true for reproducible results.

python predict.py -c /path/to/inference_config.json

Run with Docker

You can also run the script with Docker. First, build the Docker image with the following command.

cd ChemGLaM_huggingface
docker build --no-cache -t chemglam .

After building the Docker image, you can run the script with the following command. You can replace the python train.py -c config/benchmark/bindingdb_cv0.json with the script you want to run. Here is an example of how to execute the training script for a specific configuration:

docker run --gpus all -it --rm -u `id -u`:`id -g` \
-v $(pwd):/workspace chemglam \
python train.py -c config/benchmark/bindingdb_cv0.json

Contact

• Takuto Koyama: koyama.takuto.82j[at]st.kyoto-u.ac.jp

Reference

bioRxiv

@article{koyama2024chemglam,
  title={ChemGLaM: Chemical Genomics Language Models for Compound-Protein Interaction Prediction},
  author={Koyama, Takuto and Tsumura, Hayato and Matsumoto, Shigeyuki and Okita, Ryunosuke and Kojima, Ryosuke and Okuno, Yasushi},
  journal={bioRxiv},
  pages={2024--02},
  year={2024},
  publisher={Cold Spring Harbor Laboratory}
}