conplex_dev's Introduction

Adapting Protein Language Models for Rapid DTI Prediction

This repository documents the code used to generate the results for our PNAS article. The updated package, which is continuously being developed, can be found at this repository. Please submit an issue or email [email protected] with any questions.

Sample Usage

python train_DTI.py --exp-id ExperimentName --config configs/default_config.yaml

Repository Organization

src: Python files containing protein and molecular featurizers, prediction architectures, and data loading
scripts: Bash files to run benchmarking tasks
- CMD_BENCHMARK_DAVIS.sh -- Run DTI classification benchmarks on DAVIS data set. Can be easily modified for other classification data sets
- CMD_BENCHMARK_TDC_DTI_DG.sh -- Run benchmarks for TDC DTI-DG regression task
- CMD_BENCHMARK_DUDE_CROSSTYPE.sh -- Evaluate trained model on DUDe decoy performance for kinase and GPCR targets
- CMD_BENCHMARK_DUDE_WITHINTYPE.sh -- The same as above, but with half of kinase, gpcr, protease, and nuclear targets
models: Pre-trained protein language models
dataset: Data sets to benchmark on, most are from MolTrans
- DAVIS
- BindingDB
- BIOSNAP
- DUDe
nb: Jupyter notebooks for data generation and exploration
train_DTI.py -- Main training script to run DTI classification benchmarks
DUDE_evaluate_decoys.py -- Compare predictions of a trained model between a target and known true binders/decoys. Visualize embedding space
DUDE_summarize_decoys.py -- Given a directory of protein targets, summarize active/decoy discriminative performance by target type