Transformation and model choice for RNA-seq co-expression analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Existing clustering algorithms for microarray data lack clear application to RNA sequencing (RNA-seq) data.
• RNA-seq data exhibits unique characteristics, such as per-sample correlation structures, that require specialized analytical approaches.
• The identification of co-expressed genes is crucial for understanding gene function and biological pathways.

Purpose of the Study:

• To investigate the application of data transformations with Gaussian mixture models for RNA-seq co-expression analysis.
• To develop a penalized model selection criterion for choosing optimal data transformations and the number of gene clusters.
• To provide a robust statistical framework for analyzing RNA-seq co-expression patterns.

Main Methods:

• Utilized Gaussian mixture models combined with data transformations to analyze RNA-seq co-expression.
• Employed a penalized model selection criterion to objectively determine the best data transformation and cluster count.
• Applied the method to four diverse RNA-seq datasets to demonstrate its efficacy.

Main Results:

• The proposed approach successfully identified co-expressed gene clusters in RNA-seq data.
• The model selection criterion effectively chose appropriate data transformations and the number of clusters.
• The method demonstrated the ability to account for complex correlation structures within RNA-seq samples.

Conclusions:

• Data transformations and Gaussian mixture models offer a powerful framework for RNA-seq co-expression analysis.
• The developed penalized model selection criterion provides an objective means for optimizing analysis parameters.
• The `coseq` Bioconductor package facilitates the implementation and visualization of these RNA-seq co-expression analyses.