Predicting drug-target interactions and binding affinity using an optimized deep learning approach.

Area of Science:

• Computational chemistry and bioinformatics
• Artificial intelligence in drug discovery

Background:

• Accurate prediction of Drug-Target Interactions (DTIs) and Drug-Target Affinity (DTA) is vital for efficient drug discovery and repurposing.
• Conventional deep learning models often fail to capture crucial local biochemical and global structural dependencies between drugs and proteins.

Purpose of the Study:

• To introduce a novel deep learning model, CNN based Dual Attention (nCNN-DA), for enhanced prediction of DTIs and DTA.
• To improve the representational power of features extracted from drug SMILES and protein sequences.

Main Methods:

• Developed nCNN-DA, integrating 1D convolutional feature extraction with channel and spatial attention mechanisms.
• Trained and evaluated the model on three benchmark datasets: KIBA, Davis, and BindingDB.
• Assessed performance using metrics including AUPR, AUROC, MSE, Pearson correlation, and accuracy.

Main Results:

• nCNN-DA demonstrated significant performance improvements over established models like FusionNet, GraphormerDTI, DeepDTAGen, and DTBA-net.
• Achieved top accuracy rates of 98.5% (KIBA), 95.5% (Davis), and 97.5% (BindingDB).
• Reported lowest Mean Squared Error (MSE) values: 0.1559 (KIBA), 0.3189 (Davis), and 0.2957 (BindingDB), alongside superior AUPR and Pearson Correlation scores.

Conclusions:

• nCNN-DA effectively identifies putative DTI pairs and predicts binding affinities with high accuracy.
• The model's versatility and generalizability make it a valuable tool for drug discovery, virtual screening, and drug repurposing.