multiClust: An R-package for Identifying Biologically Relevant Clusters in Cancer Transcriptome Profiles

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clustering transcriptomics profiles is crucial for identifying patient subgroups.
• Gene selection is a critical step in clustering but choosing methods is challenging.
• Existing R packages lack comprehensive support for diverse feature selection methods.

Purpose of the Study:

• To develop an integrative R-package, multiClust, for easy experimentation with gene selection and clustering combinations.
• To identify the optimal clustering methodology for clinical outcome analysis using transcriptomics data.

Main Methods:

• Development of the multiClust R-package integrating various gene selection and clustering algorithms.
• Application of multiClust to transcriptomics datasets to evaluate different methodological combinations.
• Analysis of clinical outcomes based on identified patient subgroups.

Main Results:

• The multiClust package facilitates seamless integration and comparison of gene selection and clustering techniques.
• Variance-based gene ranking demonstrated strong performance across multiple datasets for clinical outcome prediction.
• Optimal gene selection requires careful consideration of the number of genes, with simple methods often being effective.

Conclusions:

• multiClust provides a valuable tool for researchers exploring gene selection and clustering in transcriptomics.
• Simple gene selection methods, like variance-based ranking, are effective for clinical outcome studies when the gene count is appropriate.
• No single gene selection or clustering methodology is universally superior; method choice depends on the specific study context.