Prediction of drug-target binding affinity using similarity-based convolutional neural network

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug-target interactions (DTIs) are vital for drug discovery.
• Current computational methods often predict binary interactions, not binding affinity strength.
• Accurate binding affinity prediction is more meaningful for drug efficacy.

Purpose of the Study:

• To develop an advanced computational model for predicting drug-target binding affinities.
• To leverage deep learning techniques with recent affinity data.
• To introduce a novel similarity-based approach for binding affinity prediction.

Main Methods:

• A similarity-based model utilizing 2-dimensional (2D) convolutional neural networks (CNNs).
• Application of 2D CNN to outer products of drug and target similarity matrices.
• Utilizing publicly available drug-target affinity data for training and validation.

Main Results:

• The proposed 2D CNN model effectively predicts drug-target binding affinities.
• Validation on multiple public datasets confirms the model's efficacy.
• This represents the first use of 2D CNN for similarity-based binding affinity prediction.

Conclusions:

• The developed model offers an effective approach for predicting drug-target binding affinities.
• This method can significantly aid the drug development process.
• Advances in deep learning applied to affinity data enhance predictive capabilities.