TC-DTA: Predicting Drug-Target Binding Affinity With Transformer and Convolutional Neural Networks

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Drug-target interactions (DTIs) are crucial for drug discovery.
• Traditional DTI prediction is often binary, but drug-target affinity (DTA) offers more detailed insights.
• Accurate DTA prediction can significantly accelerate virtual screening and drug development.

Purpose of the Study:

• To introduce TC-DTA, a novel deep learning model for drug-target affinity (DTA) prediction.
• To leverage convolutional neural networks (CNN) and transformer encoder modules for enhanced feature extraction.
• To improve the accuracy and efficiency of predicting drug-target binding strength.

Main Methods:

• Utilized deep learning, combining CNNs and transformer encoder modules.
• Processed raw drug SMILES strings and protein amino acid sequences through various encoding methods.
• Extracted features using CNNs for drugs and transformer encoders for proteins, followed by a multi-layer perceptron for affinity prediction.

Main Results:

• TC-DTA demonstrated superior performance over baseline methods (KronRLS, SimBoost, DeepDTA) on Davis and KIBA datasets.
• Achieved high accuracy in predicting binding affinity scores, validated by metrics like MSE, CI, and r_m².
• Highlighted the effectiveness of CNNs and transformer encoders in extracting meaningful sequence representations for DTA.

Conclusions:

• The TC-DTA model significantly enhances drug-target affinity prediction accuracy.
• Deep learning approaches, particularly using CNNs and transformer architectures, offer a more effective and efficient alternative to traditional methods in drug discovery.
• TC-DTA can accelerate the identification of high-affinity drug candidates, streamlining the drug development pipeline.