MSCMF-DTB: a multi-scale cross-modal fusion framework for drug-target binding prediction

Area of Science:

• Computational drug discovery
• Bioinformatics
• Machine learning in pharmacology

Background:

• Predicting drug-target binding is crucial but challenging due to complex molecular and sequence data.
• Existing models struggle to integrate diverse data types like molecular topology, chemical substructures, and protein sequences.
• Accurate prediction is essential for accelerating the identification of novel therapeutics.

Purpose of the Study:

• To develop an end-to-end deep learning framework, MSCMF-DTB, for both drug-target interaction (DTI) classification and drug-target affinity (DTA) regression.
• To effectively integrate molecular graph, chemical substructure, and protein sequence information for improved prediction accuracy.
• To provide a versatile tool for computational drug discovery, virtual screening, and drug repurposing.

Main Methods:

• MSCMF-DTB utilizes DenseGCN and a fingerprint channel for drug encoding, and TAPE-BERT with 1D CNN for protein sequence analysis.
• A cross-attention mechanism and tensor network model cross-modal drug-protein relationships for higher-order feature interaction.
• The fused representations are processed by a Multi-Layer Perceptron (MLP) for final DTI and DTA predictions.

Main Results:

• MSCMF-DTB achieved competitive and consistent performance on multiple DTI (Human, C. elegans, GPCR, BioSNAP, DrugBank) and DTA (DAVIS, KIBA) datasets.
• On the DrugBank dataset, MSCMF-DTB improved AUC and Recall by up to 3.2% and 6.1% respectively, outperforming DrugBAN.
• The model demonstrated stable DTA prediction on the KIBA dataset (MSE: 0.146, CI: 0.886, r2: 0.765) and identified biologically relevant interaction regions via attention analysis.

Conclusions:

• MSCMF-DTB is an effective deep learning framework for predicting drug-target interactions and affinities, integrating diverse molecular and protein data.
• The model shows significant improvements over existing methods on large-scale datasets and provides interpretable insights into drug-target binding.
• Its successful application in a cold-start case study highlights its practical utility for virtual screening and drug repurposing efforts.