MetaClustering: discovery of the different sample clusterings in gene expression data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression data analysis commonly uses sample clustering to identify biological subtypes, such as cancer subtypes.
• Existing methods may yield multiple biologically relevant clusterings depending on gene selection.
• Discovering these alternative clusterings and the genes supporting them is crucial for a comprehensive understanding of biological data.

Purpose of the Study:

• To propose a novel method, MetaClustering, for grouping genes based on the sample clustering they support.
• To enable the direct determination of different sample clusterings present within gene expression datasets.
• To identify groups of genes that are functionally related through shared clustering structures, potentially indicating non-linear relationships.

Main Methods:

• Genes are grouped according to the sample clustering they best fit.
• This approach allows for the direct identification of distinct sample clusterings.
• The method was validated on simulated data and real cancer gene expression datasets.

Main Results:

• MetaClustering successfully identified known cancer subtypes in real data that were previously undetectable using the complete gene set.
• The method demonstrated its capability to group genes exhibiting non-linear relationships, as evidenced by clustering cell-cycle genes together.
• Application to simulated data also yielded successful clustering outcomes.

Conclusions:

• MetaClustering provides a powerful approach to uncover hidden biological structures and subtypes within gene expression data.
• The method excels at identifying groups of genes with shared functional or structural relationships, including non-linear associations.
• This technique enhances the discovery of complex biological patterns, particularly in cancer research.