Classification pretraining-regression fine-tuning for compound-target interaction prediction

This repository documents the code used to generate the results for our article. The updated package, which is continuously being developed, can be found at this repository. Please submit an issue or email my19653@163.com with any questions.

cd ./src python main.py --model DeepCPI --split ck --mode pretrain-finetune

• src: Python modules containing protein and molecular featurizers, prediction architectures, and data loading scripts

• data: All training datasets

• Predict: Predicted target spectra for reproductive effect analysis

  • predict.py -- Loads trained models and performs prediction
  • Analyze.ipynb -- Statistical analysis of predicted target spectra
  • cluster.ipynb -- Clustering analysis of target spectra for highly active compounds

• interpret: Model interpretation and substructure alert identification

  • interpret.py -- Computes gradient-based importance and feature sensitivity
  • broad_alerts.py -- Identifies globally important substructures
  • enrichment_test.py -- Performs enrichment analysis to evaluate the significance of global substructures

• disease: Disease–target association analysis

  • Analyze.ipynb -- Computes shared and specific targets and performs disease clustering