Classification prediction of drug target binding affinity based on the MolrProtTrans model

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Accurate prediction of drug-target interactions is crucial for efficient virtual drug screening.
• Existing models often neglect protein characteristics, leading to biased predictions and suboptimal performance, especially for complex targets like G protein-coupled receptors (GPCRs).
• Transformer-based models like TransformerCPI have shown limitations in specific tasks such as label inversion on GPCR datasets.

Purpose of the Study:

• To develop an enhanced transformer-based deep learning model for predicting drug-target interactions.
• To improve prediction accuracy by integrating both molecular and protein features.
• To address the limitations of existing models in handling GPCR datasets and enhance overall virtual drug screening capabilities.

Main Methods:

• Developed an enhanced transformer-based model incorporating Molr and ProtTrans networks for feature extraction.
• Integrated a transposed attention mechanism to better capture feature relationships.
• Employed a triple-loss self-supervised learning approach to improve model robustness and accuracy.
• Evaluated model performance on GPCR label-inversion and human target datasets.

Main Results:

• The proposed model achieved an Area Under the Curve (AUC) of 0.81 on the GPCR label-inversion dataset.
• The model obtained an AUC of 0.92 on a human target dataset.
• Performance metrics were numerically higher compared to TransformerCPI and other baseline methods in experimental evaluations.

Conclusions:

• The enhanced transformer model demonstrates improved accuracy in predicting drug-target interactions.
• The integration of molecular and protein information, along with advanced deep learning techniques, significantly boosts predictive performance.
• The model shows promising potential for advancing virtual drug screening and accelerating drug discovery pipelines.