Bayesian hierarchical clustering for studying cancer gene expression data with unknown statistics

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clustering analysis is crucial for interpreting gene expression data.
• Bayesian hierarchical clustering (BHC) offers automated cluster number inference and improved quality via Bayesian model selection.

Purpose of the Study:

• To introduce Gaussian Bayesian Hierarchical Clustering (GBHC), an extension of BHC for gene expression data.
• To evaluate GBHC's performance in sample and gene clustering against established algorithms.

Main Methods:

• Representing gene expression data as Gaussian mixture models.
• Utilizing normal-gamma distribution as a conjugate prior for mean and precision in Gaussian components.
• Testing GBHC on 11 cancer and 3 synthetic datasets.

Main Results:

• GBHC demonstrated superior concordance with ground truth in sample clustering compared to other algorithms on cancer datasets.
• GBHC frequently inferred cluster numbers close to the actual number.
• GBHC produced more biologically plausible gene clustering results than state-of-the-art methods.

Conclusions:

• GBHC offers a robust and accurate alternative for gene expression data analysis.
• The algorithm shows promise for both sample and gene clustering in cancer research.
• GBHC provides improved biological interpretability in gene expression studies.