Drug-target binding affinity prediction using message passing neural network and self supervised learning.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Drug-target binding affinity (DTA) prediction is crucial for accelerating drug discovery.
• Deep learning models offer promising alternatives to traditional methods for DTA prediction.
• Existing deep learning approaches have limitations in molecular representation and protein embedding.

Purpose of the Study:

• To develop an advanced deep learning model for enhanced DTA prediction.
• To improve molecular graph representation and incorporate self-supervised learning for protein embedding.
• To provide a novel strategy for deep learning-based virtual screening.

Main Methods:

• Utilized an undirected-CMPNN (Convolutional Message Passing Neural Network) for molecular embedding.
• Integrated CPCProt (Context-Predictive learning for Proteins) and MLM (Masked Language Model) for protein embedding.
• Incorporated an attention mechanism to identify critical regions within protein sequences.

Main Results:

• The proposed model demonstrated superior performance on the Ki and Davis datasets compared to existing deep learning methods.
• The enhanced molecular and protein representations led to improved DTA prediction accuracy.
• The attention mechanism effectively highlighted important protein sequence features.

Conclusions:

• The developed model significantly improves DTA prediction performance.
• This work offers a novel and effective strategy for deep learning-based virtual screening in drug discovery.
• The findings contribute to the advancement of computational methods in pharmaceutical research.