Influence of the go-based semantic similarity measures in multi-objective gene clustering algorithm performance

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene similarity measures, particularly those derived from Gene Ontology (GO), are crucial for effective multi-objective gene clustering.
• Integrating biological knowledge enhances gene clustering by considering both co-expression and biological homogeneity.
• Selecting the appropriate GO-based semantic similarity measure is key to achieving meaningful clustering results.

Purpose of the Study:

• To investigate the influence of four prominent GO-based semantic similarity measures on multi-objective gene clustering performance.
• To determine which GO similarity measure best balances gene co-expression and biological homogeneity in clustering outcomes.
• To evaluate the comparative performance of Jiang-Conrath, Wang, Resnik, and other GO similarity measures in gene clustering.

Main Methods:

• Utilized four publicly available gene expression datasets.
• Applied a multi-objective gene clustering algorithm incorporating four GO-based semantic similarity measures: Jiang-Conrath, Wang, Resnik, and another (unspecified).
• Conducted comparative analyses using multi-objective optimization metrics and clustering performance indices.

Main Results:

• Jiang-Conrath and Wang similarities generally performed well on multi-objective metrics.
• Resnik similarity excelled in achieving high compactness and separation, indicating better gene co-expression.
• Wang similarity demonstrated superior performance in biological homogeneity, identifying more significant GO terms.

Conclusions:

• While specific GO-based semantic similarity measures show strengths in certain aspects (e.g., co-expression, homogeneity), no single measure consistently and significantly improved the overall performance of the multi-objective gene clustering algorithm.
• The choice of GO similarity measure impacts different clustering properties, suggesting a trade-off between optimizing for co-expression versus biological homogeneity.
• Further research may be needed to develop hybrid approaches or novel similarity measures that can better integrate diverse biological knowledge for improved gene clustering.