Graph-Based Clustering of Predicted Ligand-Binding Pockets on Protein Surfaces

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Identifying ligand binding pockets on protein surfaces is crucial for drug discovery.
• Existing software like PASS and Fpocket can identify multiple overlapping pockets for a single ligand.
• Representing a small molecule with a merged set of contacting pockets is desirable.

Purpose of the Study:

• To evaluate clustering approaches for merging overlapping ligand binding pockets.
• To determine the effectiveness of different algorithms in consolidating pocket predictions.

Main Methods:

• Tested three clustering approaches: classical clustering, hierarchical clustering, and a graph theory-based method using maximum flow.
• Applied these methods to pocket sets generated by PASS and Fpocket software.
• Evaluated the ability of each method to merge overlapping pockets.

Main Results:

• Hierarchical clustering and the maximum flow algorithm demonstrated effectiveness in merging pockets.
• Both methods showed favorable performance in clustering pockets predicted by PASS and Fpocket.
• These approaches can consolidate multiple predicted pockets into a more representative binding site.

Conclusions:

• Hierarchical clustering and maximum flow are suitable methods for merging overlapping ligand binding pockets.
• These computational approaches enhance the representation of small molecules in drug discovery.
• Improved pocket detection aids in identifying potential drug candidates.