Improving drug-target affinity prediction by adaptive self-supervised learning.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-target affinity prediction is crucial for efficient drug screening and discovery.
• Existing self-supervised learning methods struggle with sample mismatch and the induced-fit principle in drug-target interactions.
• These challenges limit the accuracy of computational drug-target affinity prediction.

Purpose of the Study:

• To develop an advanced computational method for drug-target affinity prediction.
• To overcome the limitations of current self-supervised learning approaches in this domain.
• To enhance the accuracy and reliability of predicting drug-target interactions.

Main Methods:

• Introduced an adaptive self-supervised learning-based drug-target affinity prediction (ASSLDTA) model.
• Integrated a novel adaptive self-supervised learning (ASSL) module for low-level feature extraction from unlabeled data.
• Employed a high-level feature learning network for precise affinity prediction using labeled data.

Main Results:

• ASSLDTA effectively bridges the objective gap and alleviates sample mismatch issues.
• The model significantly improves drug-target affinity prediction accuracy compared to existing deep learning methods.
• Learned adaptive self-supervised learning-based features demonstrated superior performance.

Conclusions:

• ASSLDTA provides a more accurate and comprehensive solution for drug-target affinity prediction.
• The two-stage feature extraction design effectively leverages different data sources and model advantages.
• The proposed method validates the effectiveness of adaptive self-supervised learning in computational drug discovery.