Q-BAFNet: A Hybrid Quantum Classical Approach for Drug-Target Binding Affinity Prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Quantum computing

Background:

• Accurate drug-target binding affinity (DTA) prediction is crucial for accelerating drug discovery and high-throughput screening.
• Existing deep learning models struggle to capture complex, context-dependent ligand-protein interactions.

Purpose of the Study:

• To develop a novel hybrid quantum-classical deep learning architecture, Q-BAFNet, for enhanced DTA prediction.
• To improve the accuracy and generalizability of DTA prediction models, especially in data-limited or biologically diverse settings.

Main Methods:

• Q-BAFNet integrates semantic, structural, and sequential molecular representations using ChemBERTa, ProtT5, and graph convolutional networks (GCNs).
• A cross-modal attention fusion mechanism dynamically aligns ligand and protein substructures.
• A variable quantum circuit (VQC) is employed to capture nonlinear and entangled dependencies in quantum Hilbert space.

Main Results:

• Q-BAFNet demonstrated superior performance across benchmark datasets (Davis, KIBA, Metz) compared to existing methods.
• The model achieved improved metrics including mean squared error (MSE), Pearson correlation coefficient (PCC), concordance index (CI), and R-squared (R²).
• Exceptional performance was observed in zero-shot prediction scenarios, including drug and target cold-start evaluations.

Conclusions:

• Q-BAFNet represents a significant advancement in DTA prediction by effectively leveraging hybrid quantum-classical deep learning.
• The model's ability to capture complex molecular interactions offers a promising avenue for robust and generalizable predictions in drug discovery.