Fusing Sequence and Structural Knowledge by Heterogeneous Models to Accurately and Interpretively Predict Drug-Target...

Area of Science:

• Computational Chemistry
• Bioinformatics
• Drug Discovery

Background:

• Drug-target affinity (DTA) prediction is vital for drug discovery, but current computational methods lack predictive accuracy.
• Existing models often fail to capture the intricate structural details of molecular interactions.

Purpose of the Study:

• To develop a novel deep learning model, S2DTA, for accurate DTA prediction.
• To leverage fused sequence and graph structural features using heterogeneous models.

Main Methods:

• Developed S2DTA, a deep learning model fusing drug SMILES, target, and pocket sequence features with graph structural information.
• Employed heterogeneous models based on graph and semantic networks for feature integration.
• Utilized a bidirectional self-attention mechanism for interpretability analysis.

Main Results:

• S2DTA demonstrated superior predictive accuracy compared to DeepDTA, GraphDTA, and DeepDTAF.
• Achieved a 25.2% reduction in Mean Absolute Error (MAE) and a 20.1% decrease in Root Mean Square Error (RMSE).
• Showcased significant improvements in Pearson Correlation Coefficient (PCC), Spearman, Concordance Index (CI), and R-squared (R²).

Conclusions:

• The integration of heterogeneous models significantly boosts predictive accuracy in DTA prediction.
• S2DTA is an effective and accurate tool for predicting drug-target affinity, advancing drug discovery efforts.
• Interpretability analysis confirmed the model's reliability.