Fourier-Enhanced Kolmogorov-Arnold Network With Attention for Drug-Target Interaction Prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Identifying drug-target interactions (DTI) is crucial but expensive in drug discovery.
• Developing accurate and efficient computational methods for DTI prediction is essential.

Purpose of the Study:

• To propose FKAN-a, a novel framework for DTI prediction.
• To integrate Fourier-enhanced Kolmogorov-Arnold networks (KAN) with attention mechanisms and contrastive learning.

Main Methods:

• Preprocess and encode drug molecules and protein sequences using pretrained representations.
• Utilize KAN with learnable Fourier bases to model complex nonlinear relationships.
• Employ a cross-modality attention module for fine-grained drug-protein association modeling.

Main Results:

• FKAN-a demonstrates superior prediction performance compared to state-of-the-art methods.
• The framework shows enhanced computational efficiency.
• Consistent outperformance across three public benchmark datasets was observed.

Conclusions:

• FKAN-a offers an effective computational solution for DTI prediction.
• The proposed framework aids in practical candidate prioritization for drug discovery.