KASSPer: kinase active site structure prediction using protein and ligand language models and its application to...

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Structure-based virtual screening (SBVS) is hindered by the rigid-receptor assumption, especially for kinases with multiple active-site conformations.
• Existing ensemble screening methods are computationally expensive and may not fully capture dynamic receptor states.

Purpose of the Study:

• To introduce KASSPer (Kinase Active Site Structure Predictor), a novel framework for predicting kinase active-site conformational states.
• To enable ligand-specific conformer selection for structure-based virtual screening, reducing computational costs.

Main Methods:

• Utilized protein and compound language models to predict kinase active-site conformational states.
• Developed a framework that takes kinase amino acid sequence and ligand SMILES string as input.
• Integrated KASSPer for ligand-specific conformer selection prior to structure-based virtual screening.

Main Results:

• KASSPer successfully predicts kinase active-site conformational states.
• KASSPer-guided screening demonstrated superior performance compared to ensemble-based approaches on the DUD-E kinase subset.
• The framework offers a computationally efficient alternative to exhaustive ensemble screening.

Conclusions:

• KASSPer enhances structure-based virtual screening by accurately predicting relevant kinase conformations.
• The developed framework offers a significant advancement in computationally efficient drug discovery for kinases.
• KASSPer-guided screening provides a more effective approach for identifying potential kinase inhibitors.