Bidirectional cross-modal fusion with tensor interaction for drug-target binding prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Artificial intelligence in drug discovery

Background:

• Accurate drug-target binding affinity prediction is crucial for computational drug discovery.
• Current deep learning methods often isolate drug and protein representations, limiting cross-modal interaction capture.
• Complex, non-linear interactions govern binding, posing challenges for existing models.

Purpose of the Study:

• To propose a novel framework, Bidirectional Cross-Modal Fusion with Tensor Interaction (BiT-Fusion), to enhance drug-target binding affinity prediction.
• To improve the modeling of cross-modal dependencies between drug molecules and protein targets.
• To enable more effective information exchange between molecular graphs and protein sequences.

Main Methods:

• Developed BiT-Fusion, a framework utilizing bidirectional fusion and tensor interaction.
• Strengthened interaction modeling by integrating drug and protein features early and bidirectionally.
• Employed multiplicative coupling between drug and protein features to capture complex relationships.

Main Results:

• BiT-Fusion demonstrated competitive and consistent improvements on the Davis and KIBA benchmarks.
• The framework achieved enhanced performance across multiple evaluation metrics.
• Ablation studies confirmed bidirectional fusion and tensor interaction as key performance drivers.

Conclusions:

• Enhancing cross-modal interaction learning is a practical approach for improving drug-target binding prediction.
• BiT-Fusion offers a more effective way to model interactions for AI-enabled drug discovery.
• The findings have implications for precision medicine and global health initiatives (SDG 3).