CAVIAR: a method for automatic cavity detection, description and decomposition into subcavities.

Area of Science:

• Structural Biology
• Computational Chemistry
• Bioinformatics

Background:

• Accurate protein binding site description is crucial for determining similarity and applying machine learning to functional predictions.
• Existing methods may lack comprehensive descriptor generation for diverse structural inputs.

Purpose of the Study:

• Introduce CAVIAR, an open-source tool for generating novel descriptors for protein binding sites.
• Demonstrate the utility of CAVIAR descriptors in machine learning tasks, specifically predicting binding site ligandability.
• Enable automatic identification and characterization of subcavities within binding sites.

Main Methods:

• Utilized protein structures in PDB and mmCIF formats, along with molecular dynamics simulation trajectory frames as input.
• Developed CAVIAR to generate descriptors capturing geometric and chemical properties of binding sites.
• Implemented algorithms for automatic subcavity and subpocket assignment.

Main Results:

• CAVIAR descriptors were successfully applied to predict binding site ligandability using machine learning.
• The tool automatically assigns subcavities, mimicking medicinal chemistry empirical definitions.
• Experimental binding affinity correlated with the number of occupied subcavities, with >3 subcavities yielding nanomolar affinities.

Conclusions:

• CAVIAR provides a robust method for generating binding site descriptors applicable to machine learning.
• The tool facilitates protein engineering and hit identification by enhancing binding site analysis.
• Open-source availability on GitHub and Anaconda cloud promotes widespread adoption and further research.