DCGAN-DTA: Predicting drug-target binding affinity with deep convolutional generative adversarial networks.

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning

Background:

• Computational methods are increasingly used to predict drug-target binding affinity, accelerating drug discovery.
• Existing machine learning methods face challenges like limited data, manual feature engineering, and insufficient validation.
• Experimental methods for binding affinity prediction are costly and time-consuming.

Purpose of the Study:

• To develop a novel deep convolutional generative adversarial network (DCGAN) for accurate drug-target binding affinity prediction.
• To address limitations of current computational and experimental drug discovery approaches.
• To provide a robust and validated computational tool for drug discovery.

Main Methods:

• Utilized deep convolutional generative adversarial networks (DCGANs) for binding affinity prediction.
• Conducted validation and adversarial control experiments using straw models.
• Evaluated the method on BindingDB and PDBBind datasets using warm-start and cold-start settings.

Main Results:

• The proposed DCGAN method demonstrated superior predictive performance compared to baseline and state-of-the-art methods.
• Outperformed alternatives in three performance measures under warm-start conditions.
• Showed enhanced predictive accuracy, especially in concordance index, for physiochemical-based cold-start settings.

Conclusions:

• The DCGAN-based method offers practical value and superior performance for drug-target binding affinity prediction.
• This approach can accelerate drug repurposing, novel drug discovery, and disease treatment.
• Source code and a web server are available for broader application and validation.