ClusterMine: A knowledge-integrated clustering approach based on expression profiles of gene sets

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data clustering is crucial for identifying cell subpopulations and disease subtypes.
• Conventional methods like hierarchical clustering (HC) and consensus clustering (CC) use holistic gene profiles, limiting biological interpretability.
• These methods often fail to identify specific gene sets contributing to cluster formation.

Purpose of the Study:

• To introduce ClusterMine, a novel approach for clustering gene expression data.
• To integrate prior knowledge of annotated gene sets for improved clustering performance and biological interpretability.
• To enable the identification of gene sets that drive sample clustering.

Main Methods:

• ClusterMine assesses functional similarity between samples by integrating annotated gene sets from databases like Gene Ontology.
• It provides cluster membership for each sample, similar to conventional approaches.
• The method outputs the most relevant gene sets contributing to the identified clusters.

Main Results:

• ClusterMine demonstrated superior performance compared to conventional clustering methods across nine diverse biological datasets.
• The gene sets prioritized by ClusterMine were found to be biologically meaningful and interpretable.
• The approach successfully facilitated the biological interpretation of identified sample clusters.

Conclusions:

• ClusterMine offers a significant advancement in gene expression data analysis by enhancing clustering interpretability.
• Integrating functional gene set knowledge provides deeper biological insights into sample groupings.
• ClusterMine, available as an R package, facilitates the discovery of biologically relevant gene sets driving specific clusters.