BPCI-DTI: Drug-Target Interaction Prediction via Structure-Sensitive Drug Graph Feature Extraction and Cross-Modal Interaction Fusion

Le résumé fourni par la source

Drug-target interaction (DTI) prediction is an important stage in drug research. Accurate prediction relies on collecting drug and target properties and understanding their complicated interactions. However, conventional approaches struggle to capture small structural differences in medicinal compounds and fail to extract long-range dependence in protein sequences. Moreover, drug and protein characteristics are often combined through simple concatenation, overlooking cross-modal complementarity. To address these constraints, we present a novel DTI prediction model, BPCI-DTI. The model presents a biomolecular graph extractor, BioGraphNet, which leverages multi-level graph isomorphism networks to learn local neighborhood and global topological properties of medicinal compounds, boosting sensitivity to structural alterations. It also blends convolutional neural networks and multi-head convolutional selfattention to capture both local and long-range relationships in protein sequences. Additionally, a cross-modal interaction fusion module with a gating mechanism successfully gathers drug-target interaction information for better prediction accuracy. BPCIDTI was examined on four DTI datasets and outperformed six baseline approaches in AUROC and AUPRC, exhibiting robust predictive performance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.