Multi-way overlapping clustering by Bayesian tensor decomposition.

Area of Science:

• Genomics
• Statistical Genetics
• Bioinformatics

Background:

• Modern sequencing technologies enable large-scale gene expression analysis across diverse tissues and individuals.
• Analyzing three-way variation (genes, tissues, individuals) presents significant statistical inference challenges.
• Understanding gene expression patterns is crucial for identifying disease-related genetic factors.

Purpose of the Study:

• To develop a Bayesian multi-way clustering approach for simultaneous clustering of genes, tissues, and individuals.
• To automatically determine the optimal number of clusters using a Bayesian nonparametric prior.
• To apply the method to RNA-sequencing data for biological discovery, specifically in relation to depression.

Main Methods:

• A Bayesian hierarchical model is proposed to adaptively trichotomize data into latent categories.
• Latent variables are decomposed into lower-dimensional features representing overlapping clusters.
• The Indian buffet process is utilized as a Bayesian nonparametric prior for automatic cluster number determination.

Main Results:

• Simulation studies demonstrate the utility and performance of the proposed clustering approach.
• Application to Genotype-Tissue Expression (GTEx) RNA-seq data reveals significant findings.
• Identified depression-related genes in the human brain, consistent with existing biological knowledge.

Conclusions:

• The Bayesian multi-way clustering method effectively handles complex, multi-dimensional gene expression data.
• The approach facilitates the discovery of biologically relevant gene clusters and their associations.
• Findings on depression-related genes highlight the method's potential for advancing psychiatric genetics research.