ProtFPreDTI: Drug-Target Interaction Prediction Study and LIME Interpretability Analysis Based on the ProtBERT Deep...

Area of Science:

• Computational chemistry and cheminformatics
• Bioinformatics and computational biology
• Machine learning in drug discovery

Background:

• Drug-target interaction (DTI) analysis is crucial for pharmaceutical research and development.
• Traditional experimental methods for DTI analysis are time-consuming and expensive.
• Existing computational models face challenges with feature characterization and data imbalance.

Purpose of the Study:

• To develop a novel machine learning-based prediction method for drug-target interactions.
• To address limitations in feature extraction, data imbalance, and model interpretability in DTI prediction.
• To improve the efficiency and accuracy of drug discovery processes.

Main Methods:

• Utilized Mol2Vec for drug molecule feature extraction and ProtBERT for protein sequence feature extraction.
• Employed SHAP value analysis for quantitative feature importance screening, retaining 300 dimensions.
• Implemented a fuzzy logic-based undersampling strategy for data balancing and an adaptive weighted fusion of XGBoost and random forest for prediction.
• Integrated LIME for model interpretability.

Main Results:

• The ProtFPreDTI model achieved an Area Under the Curve (AUC) of 0.92 in independent validation.
• Demonstrated significant improvements in prediction accuracy, sensitivity, and specificity compared to traditional methods.
• The system showed enhanced prediction robustness and cross-dataset generalization capabilities.

Conclusions:

• The developed ProtFPreDTI model offers a robust and accurate solution for predicting drug-target interactions.
• The methodology optimizes the entire process from feature engineering to result interpretation, enhancing drug discovery efficiency.
• The approach provides a scientific and traceable decision-making basis for drug development.