SSR-DTA: Substructure-aware multi-layer graph neural networks for drug-target binding affinity prediction.

Area of Science:

• Computational chemistry and drug discovery.
• Artificial intelligence in cheminformatics.
• Bioinformatics and structural biology.

Background:

• Accurate drug-target binding affinity (DTA) prediction is crucial for efficient drug discovery.
• Current graph neural networks (GNNs) struggle with extracting substructural features of varying sizes and integrating protein structural information.
• Sequence-based models for protein targets lack essential structural context.

Purpose of the Study:

• To develop an advanced AI model, SSR-DTA, for more accurate DTA prediction.
• To overcome limitations in feature extraction across diverse molecular scales and integrate multi-modal protein data.
• To improve the robustness and accuracy of DTA predictions in drug discovery pipelines.

Main Methods:

• Introduction of SSR-DTA, a multi-layer graph network designed for adaptable feature extraction across different molecular scales.
• Integration of BiGNN to simultaneously process protein sequence (primary structure) and graph-based structural (tertiary structure) information.
• Rigorous experimental validation on four benchmark DTA datasets.

Main Results:

• SSR-DTA demonstrates superior performance compared to existing state-of-the-art models.
• Achieved a 20% reduction in mean squared error on the Davis dataset and a 5% reduction on the KIBA dataset.
• The model effectively captures richer biological features and integrates sequence and structural data for enhanced prediction.

Conclusions:

• SSR-DTA offers a robust and accurate approach to DTA prediction by addressing limitations of previous GNNs and sequence-based methods.
• The model's ability to handle diverse structural sizes and integrate multi-modal data makes it a valuable tool for accelerating drug discovery.
• SSR-DTA shows significant potential for reducing labor and financial losses by improving the screening of effective drug candidates.