MFCADTI

MFCADTI: improving drug-target interaction prediction by integrating multiple feature through cross attention mechanism

Fig. 1. The framework of MFCADTI.The MFCADTI framework consists of three modules: A. Network feature extraction. B. Attribute feature extraction. C. Cross-attention feature fusion and prediction.

Dependencies

• pytorch==1.12.0
• numpy==1.23.1
• pandas==1.4.3
• tensorboardX==2.6
• tensorflow==2.6.0
• keras==2.9.0
• scikit-learn==1.1.1
• RDKit==2022.9.3
• gensim==4.2.0
• subword-nmt==0.3.8

Dataset

In the dataset folder, we provide the processed data of Luo dataset.

ESFP and dictionary

The ESFP folder contains the data needed for the FCS embedding method that is built based on MolTrans https://github.com/kexinhuang12345/MolTrans. The "dictionary" directory includes the dictionaries constructed for drug and target sequences in the luo dataset.

Resources

• main.py: train and test the model
• hyperparameter.py: set the hyperparameter of CrossAttentionDTI
• model.py: CrossAttentionDTI model architecture
• measure.py: The module for calculating metrics

Setting directory

Make the "result" directory before running the model. The run results are saved in the "result" directory

Run

python main.py