GS-DTA: integrating graph and sequence models for predicting drug-target binding affinity.

Area of Science:

• Computational Chemistry
• Bioinformatics
• Drug Discovery

Background:

• Drug-target binding affinity (DTA) prediction is crucial for drug discovery and repositioning.
• Existing methods struggle with complex molecular structures and long-range protein interactions.
• There's a need for improved DTA prediction models that capture intricate molecular and protein features.

Purpose of the Study:

• To develop a novel graph and sequence-based method, GS-DTA, for accurate DTA prediction.
• To address limitations in current models regarding the analysis of important molecular nodes and protein structural information.
• To improve the exploration of relationships within complex drug molecules and between distant amino acid fragments.

Main Methods:

• GS-DTA utilizes simplified molecular input line entry system (SMILES) for drugs and amino acid sequences for proteins.
• Drug features are extracted using Graph Attention Network version 2-Graph Convolutional Network (GATv2-GCN) and a three-layer GCN.
• Protein features are extracted using a combination of Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM), and Transformer networks.

Main Results:

• GS-DTA models drugs as graphs, with GATv2-GCN focusing on important atomic nodes and GCN capturing hierarchical features.
• The protein framework extracts comprehensive contextual and structural information from amino acid sequences.
• Drug and protein feature vectors are integrated for DTA prediction via a fully connected layer.

Conclusions:

• GS-DTA demonstrates strong performance on the Davis and KIBA datasets, indicated by favorable MSE, CI, and r²_m values.
• The method improves the accuracy of drug-target binding affinity prediction.
• GS-DTA offers a promising approach for computational drug discovery and development.