ASAP-DTA: Predicting drug-target binding affinity with adaptive structure aware networks

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Drug-target affinity (DTA) prediction is vital for drug repurposing and resource optimization.
• Current methods face challenges in effectively capturing complex molecular representations.

Purpose of the Study:

• To develop a novel graph-based deep learning model for accurate DTA prediction.
• To enhance feature extraction from molecular graphs using adaptive structure-aware pooling and self-attention.

Main Methods:

• A graph neural network integrated with a self-attention mechanism for node significance.
• Adaptive structure-aware pooling (global and hierarchical) for molecular graph processing.
• Clustering adjacent nodes and weighting features by attention scores for molecular representation.

Main Results:

• Achieved the lowest mean squared error (MSE) of 0.126 on the KIBA dataset, a 0.5% improvement over baseline.
• Demonstrated superior performance in both regression and binary classification tasks, validating generalization capabilities.

Conclusions:

• The proposed model offers a significant advancement in graph feature extraction for DTA prediction.
• The model's effectiveness and generalization potential are confirmed across benchmark datasets and task types.