BioFusionDTI: Assimilating Graph and Sequence Modalities for Generalizable Drug-Target Interaction Prediction.

Area of Science:

• Computational Biology
• Drug Discovery
• Artificial Intelligence

Background:

• Accurate drug-target interaction (DTI) prediction is crucial for efficient drug discovery and repurposing.
• Current deep learning models struggle with generalization in cold-start scenarios and lack interpretability.

Purpose of the Study:

• To develop BioFusionDTI, a multimodal deep learning framework to enhance DTI prediction accuracy and interpretability.
• To address limitations in generalization and interpretability of existing deep learning models for DTIs.

Main Methods:

• Integrating graph-based (using Graph Convolutional Networks - GCNs) and sequence-based (using Convolutional Neural Networks - CNNs) representations of drugs and proteins.
• Employing a Bilinear Attention Network (BAN) for fine-grained cross-modal interaction capture.
• Utilizing pretrained biomolecular language models for sequence embeddings.

Main Results:

• BioFusionDTI consistently outperformed state-of-the-art baselines across multiple benchmark datasets (SNAP, DRH, Kinase) in various settings (warm, cold-drug, cold-protein).
• Ablation studies confirmed the effectiveness of the fusion strategy, with the BAN module significantly contributing to performance.
• Attention visualizations identified biologically plausible interaction sites, consistent with molecular docking results.

Conclusions:

• BioFusionDTI offers a robust and interpretable solution for predicting drug-target interactions.
• The framework demonstrates improved generalization capabilities, particularly in challenging cold-start scenarios.
• BioFusionDTI advances the field of computational drug discovery and repositioning.