Exploring the Druggable Conformational Space of Protein Kinases Using AI-Generated Structures

Area of Science:

• Structural biology
• Computational chemistry
• Drug discovery

Background:

• Protein kinases play crucial roles in cellular functions and are key targets for drug development.
• Drug discovery is hindered by a lack of experimentally determined structures for kinases in inactive conformations.
• Kinase conformational states, regulated by motifs like DFG and aC-Helix, influence drug binding and therapeutic outcomes.

Approach:

• Evaluated conformational space of kinases in the Protein Data Bank (PDB) and AI-predicted models (AlphaFold2, ESMFold).
• Investigated AlphaFold2's ability to predict diverse kinase conformations by varying multiple sequence alignment (MSA) depth.
• Demonstrated that reduced MSA depth in AlphaFold2 facilitates exploration of alternative kinase conformations.

Key Points:

• Both PDB and AI-predicted kinase structures show a bias towards active conformations.
• AlphaFold2 and ESMFold models overrepresent active kinase states, similar to PDB data.
• Predicting kinase structures with AlphaFold2 at lower MSA depths reveals previously unobserved conformations for 398 kinases.

Conclusions:

• Lowering MSA depth in AlphaFold2 enables broader exploration of kinase conformational space.
• This approach can identify novel, potentially druggable kinase conformations.
• The findings offer a new strategy for discovering therapeutics targeting challenging kinase proteins.