Co-VAE: Drug-Target Binding Affinity Prediction by Co-Regularized Variational Autoencoders.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Accurate prediction of drug-target interactions is crucial for efficient drug discovery.
• Predicting drug-target binding affinity, a measure of interaction strength, is computationally challenging.
• Existing methods often focus on interaction prediction rather than affinity quantification.

Purpose of the Study:

• To develop a novel computational model for predicting drug-target binding affinity.
• To enhance drug discovery by enabling the prediction of binding strength and generation of new drug candidates.
• To improve upon existing drug-target affinity prediction techniques.

Main Methods:

• Proposed a novel co-regularized variational autoencoder (Co-VAE) model.
• Utilized two VAEs for generating drug SMILES strings and target sequences.
• Incorporated a co-regularization component for predicting binding affinities.

Main Results:

• The Co-VAE model demonstrated superior performance in predicting drug-target binding affinity compared to DeepDTA and DeepAffinity.
• The model successfully generated novel, valid drug candidates with similar targets.
• Theoretical analysis confirmed the model maximizes the joint likelihood of drug, protein, and their affinity.

Conclusions:

• The Co-VAE model offers a powerful approach for predicting drug-target binding affinity.
• This method advances drug discovery by enabling accurate affinity prediction and de novo drug generation.
• Co-VAE provides a significant improvement over existing computational methods in both prediction accuracy and drug generation capabilities.