Kinome-wide interaction modelling using alignment-based and alignment-independent approaches for kinase description...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Protein kinases are vital in cell functions; dysregulation causes diseases like cancer.
• Current kinase inhibitors lack selectivity, leading to toxic side effects.
• Bioinformatics can predict inhibitor-kinase interactions for improved drug design.

Purpose of the Study:

• To develop and validate a proteochemometric model for predicting inhibitor-kinase interactions.
• To identify optimal methods for encoding kinase properties and correlation analysis.
• To assess the model's performance and scalability.

Main Methods:

• Applied proteochemometric modeling to 317 kinases and 38 inhibitors (12,046 combinations).
• Utilized amino acid physico-chemical z-scale descriptors for kinase sequences.
• Employed support vector machines and partial least-squares for correlation analysis.
• Validated models using double cross-validation.

Main Results:

• Achieved high predictive ability for new kinase-inhibitor pairs (P2 = 0.67-0.73) and new kinases (P2kin = 0.65-0.70).
• Demonstrated high sensitivity and specificity in distinguishing interacting pairs (AUC = 0.92-0.93).
• A valid model was obtained using only 10% of the data, showing scalability.

Conclusions:

• Proteochemometrics is highly applicable for kinome-wide interaction modeling.
• This approach can accelerate the identification and optimization of kinase inhibitors.
• Enables the design of more selective kinase-targeted and multi-targeted therapies.