MREDTA: A BERT and transformer-based molecular representation encoder for predicting drug-target binding affinity.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Drug-target binding affinity (DTA) prediction is crucial for identifying new therapeutic applications of existing drugs (drug repositioning).
• Current DTA models face challenges in achieving high accuracy and generalizability.
• Developing robust models is essential for efficient drug discovery pipelines.

Purpose of the Study:

• To introduce a novel model, Molecular Representation Encoder-based DTA prediction (MREDTA), for accurate and generalizable DTA prediction.
• To leverage advanced deep learning architectures for enhanced molecular feature extraction.
• To validate the model's performance against existing state-of-the-art methods.

Main Methods:

• MREDTA integrates BERT-Trans Block, Multi-Trans Block, and DTI Learning modules.
• The model simultaneously extracts local and global molecular features using skip connections.
• The Multi-Trans Block enhances sensitivity to molecular structures, while BERT improves generalizability.

Main Results:

• MREDTA demonstrated optimal performance on the KIBA and Davis datasets, outperforming 12 advanced models.
• A case study involving 2034 FDA-approved drugs for non-small-cell lung cancer (NSCLC) targeting mutant EGFR^T790M validated MREDTA's robustness.
• Molecular docking results confirmed the model's reliability in predicting drug-target interactions.

Conclusions:

• MREDTA offers a significant advancement in DTA prediction accuracy and generalizability.
• The model shows promise for accelerating drug repositioning and personalized medicine, particularly for complex targets like mutant EGFR.
• MREDTA's architecture provides a robust framework for future development in computational drug discovery.