Kinase-inhibitor binding affinity prediction with pretrained graph encoder and language model.

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Accurate prediction of inhibitor-kinase binding affinity is critical for drug discovery and cancer treatment.
• Current prediction methods struggle with data representation, feature extraction, and capturing complex kinase-inhibitor interactions.
• Advanced artificial intelligence (AI) and deep learning techniques show promise but require further development for intricate molecular interactions.

Purpose of the Study:

• To develop a novel framework, Kinhibit, for enhanced inhibitor-kinase binding affinity prediction.
• To address limitations in existing methods, including insufficient data expression and feature extraction.
• To improve the accuracy and effectiveness of computational tools in drug screening and biological sciences.

Main Methods:

• Kinhibit integrates self-supervised graph contrastive learning for effective feature extraction.
• Multiview molecular graph representation captures diverse molecular characteristics.
• A structure-informed protein language model (ESM-S) and feature fusion optimize inhibitor-kinase interaction analysis.

Main Results:

• Kinhibit achieved 92.6% accuracy in predicting inhibitor binding for three key mitogen-activated protein kinase (MAPK) pathway kinases (RAF, MEK, ERK).
• The framework demonstrated superior performance with 92.9% accuracy on the comprehensive MAPK-All dataset.
• Experimental results validate the effectiveness of the proposed feature extraction and fusion strategies.

Conclusions:

• Kinhibit offers a significant advancement in inhibitor-kinase binding affinity prediction.
• The framework provides a promising and effective computational tool for accelerating drug screening.
• This approach enhances our ability to understand and target kinase pathways in diseases like cancer.