Graphlet kernels for prediction of functional residues in protein structures

Area of Science:

• Structural bioinformatics
• Computational biology
• Machine learning in proteomics

Background:

• Accurate functional residue annotation is crucial for understanding protein mechanisms.
• Existing methods often rely on sequence information or simplified residue microenvironments.
• Predicting functional sites, especially phosphorylation sites, remains challenging.

Purpose of the Study:

• To introduce a novel graph-based kernel method for functional residue annotation in protein structures.
• To evaluate the method's performance on catalytic residue identification and phosphorylation site prediction.
• To compare the graphlet kernel approach against sequence-based and FEATURE methods.

Main Methods:

• Representing protein structures as contact graphs with residues as nodes.
• Encoding residues using vectors of labeled non-isomorphic subgraph (graphlet) counts.
• Employing a supervised learning framework with a graphlet kernel for residue classification.

Main Results:

• The graphlet kernel method demonstrated favorable performance on both catalytic residue and phosphorylation site prediction tasks.
• The method showed a significantly larger performance margin for phosphorylation site prediction compared to other tasks.
• Evidence suggests that surface accessibility and FEATURE's microenvironment measures are insufficient for structured phosphorylation sites.

Conclusions:

• Graph-based kernel methods, specifically using graphlet representations, offer a powerful approach for functional residue annotation.
• The graphlet kernel effectively captures local connectivity patterns for improved prediction accuracy.
• This method provides a valuable tool for advancing our understanding of protein function and regulation.