Differentiating Inhibitors of Closely Related Protein Kinases with Single- or Multi-Target Activity via Explainable...

Area of Science:

• Biochemistry and Molecular Biology
• Medicinal Chemistry
• Computational Biology

Background:

• Protein kinases are critical drug targets, with inhibitors often targeting the conserved adenosine triphosphate (ATP) binding site.
• This conserved binding site leads to the common assumption that kinase inhibitors are promiscuous.
• However, existing data suggests that many kinase inhibitors exhibit selectivity.

Purpose of the Study:

• To investigate if chemical structures can differentiate inhibitors targeting closely related human kinases with single- or multi-kinase activity.
• To identify structural features responsible for the selectivity of kinase inhibitors.

Main Methods:

• Development of a test system using two distinct kinase triplets.
• Assembly of inhibitors with reported triple-kinase and corresponding single-kinase activities.
• Application of machine learning models based on chemical structure to classify inhibitors.
• Utilizing a model-independent explanatory approach to identify predictive structural features.

Main Results:

• Machine learning models accurately distinguished between multi- and single-kinase inhibitors based on chemical structure.
• Decisive structural features were identified in multi-kinase inhibitors that were absent in single-kinase inhibitors.
• These features formed coherent, chemically meaningful substructures characteristic of multi-kinase inhibitors.

Conclusions:

• Chemical structure is a key determinant in differentiating kinase inhibitors with varying selectivity profiles.
• The identified structural features provide a mechanistic rationale for the observed selectivity of kinase inhibitors.
• This study offers valuable insights for the rational design of selective kinase inhibitors for therapeutic applications.