Improving Compound-Protein Interaction Prediction by Self-Training with Augmenting Negative Samples.

Authors: Takuto Koyama, Shigeyuki Matsumoto, Hiroaki Iwata, Ryosuke Kojima, and Yasushi Okuno
Journal of Chemical Information and Modeling
https://pubs.acs.org/doi/10.1021/acs.jcim.3c00269

Installation

Dependencies can be installed with conda:

conda env create -f environment.yml
conda activate kmol
bash install.sh

Dataset available

Datsets used for this work is avaiable. (https://drive.google.com/drive/folders/1LILb8msjzWFPdM79dOWF1XZzSqerxm4d?usp=share_link)

Usage of kMoL

Please refer to the original repository for the usage of kMoL v.1.0.1 (https://github.com/elix-tech/kmol).

Directory structure

.
├── data
|   ├── self_training_kinase             : data for kinase families
|        ├──cv1
|            ├──kinase_self_training
|                ├── config/             : configuration file
|                ├── data/               : dataset (csv)
|                ├── split/              : json files for split
|                └── run.sh              : script for running self-training
├── docker/                              : 
├── src                                  : source code
│   ├── kmol/                            : modules for kmol
│   ├── mila/                            : modules for federated learning
│   └── self                             : scripts for self-training
├── LICENSE                              : LICENSE file
└── README.md                            : this file

After moving to the ./data/self_training_*/cv1/*_self_training/ directory, you can initialize the self-training by the following command:

bash run.sh