DTBA-net: Drug-Target Binding Affinity prediction using feature selection in hybrid CNN model

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate Drug-Target Binding Affinity (DTBA) prediction is crucial for virtual screening and drug repositioning.
• Challenges in DTBA prediction include limited data, high-dimensional biochemical data, and heterogeneous data sources.
• Existing deep-learning frameworks struggle with these complexities.

Purpose of the Study:

• To develop a novel hybrid neural network model, DTBA-Net, for enhanced DTBA prediction accuracy and efficiency.
• To address the limitations of current methods in handling complex biochemical data for DTBA prediction.

Main Methods:

• DTBA-Net utilizes a hybrid Convolutional Neural Network (CNN) architecture.
• Protein sequences and compound structures are processed through the CNN, incorporating convolutional layers, a flattened layer, and dense blocks.
• An optimized feature selection process using the Modified JAYA Algorithm is integrated to reduce computational overhead and improve performance.

Main Results:

• DTBA-Net achieved high accuracy on benchmark datasets, including an R-squared value of 0.95 and a Mean Absolute Error (MAE) of 0.17 on the DAVIS dataset.
• Further validation with the drug Nirmatrelvir yielded an R-squared value of 0.96, demonstrating robustness and scalability.
• The integration of optimized feature selection accelerated model training and enhanced prediction accuracy.

Conclusions:

• DTBA-Net offers a scalable, efficient, and accurate solution for DTBA prediction.
• The model's performance facilitates faster and more reliable drug discovery processes.
• DTBA-Net shows significant potential in advancing computational drug discovery methodologies.