Deep Drug-Target Binding Affinity Prediction Base on Multiple Feature Extraction and Fusion

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-target binding affinity (DTA) prediction is vital for efficient drug discovery.
• Current deep learning models face challenges in representing drug/target correlations and effectively fusing interaction data.
• Existing methods often use simple concatenation for interaction learning, limiting exploration of complex relationships.

Purpose of the Study:

• To develop an advanced deep learning model for enhanced DTA prediction.
• To address limitations in current models regarding feature representation and interaction fusion.
• To propose BTDHDTA, an end-to-end sequence-based model for improved DTA prediction.

Main Methods:

• Utilized bidirectional gated recurrent unit (GRU), transformer encoder, and dilated convolution for feature extraction, capturing global, local, and correlational patterns.
• Implemented a novel module combining convolutional neural networks (CNNs) with Highway connections for deep feature fusion.
• Developed an end-to-end sequence-based deep learning architecture named BTDHDTA.

Main Results:

• BTDHDTA demonstrated superior performance on benchmark datasets (Davis, KIBA, Metz) compared to state-of-the-art methods.
• Achieved improved metrics including Mean Squared Error (MSE), Concordance Index (CI), and R-squared (R²).
• Successfully predicted binding affinities for FDA-approved drugs against SARS-CoV-2 proteins in a case study, validating practical utility.

Conclusions:

• BTDHDTA offers a significant advancement in DTA prediction accuracy and efficiency.
• The model's ability to capture intricate data correlations and fuse features enhances predictive power.
• BTDHDTA shows promise for real-world applications, including drug repurposing and development against novel pathogens like SARS-CoV-2.