THREE-WAY CLUSTERING OF MULTI-TISSUE MULTI-INDIVIDUAL GENE EXPRESSION DATA USING SEMI-NONNEGATIVE TENSOR...

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High-throughput sequencing generates large multi-tissue gene expression datasets.
• Gene expression variation arises from complex interactions between genes, tissues, and individuals.
• Traditional clustering methods are inadequate for exploring these three-way interactions.

Purpose of the Study:

• To develop a statistical method for investigating transcriptome variation across individuals and tissues simultaneously.
• To address the limitations of classical clustering in analyzing complex, multi-way interactions in gene expression data.
• To leverage tensor decomposition for a more comprehensive analysis of multi-tissue transcriptomic data.

Main Methods:

• Proposed a novel statistical method, MultiCluster, utilizing semi-nonnegative tensor decomposition.
• Developed a tensor projection procedure for detecting covariate-related genes.
• Applied the method to GTEx RNA-seq data from 53 human tissues.

Main Results:

• MultiCluster accurately identifies three-way interactions between genes, tissues, and individuals.
• The tensor projection procedure demonstrates high power in detecting covariate-related genes.
• The method shows robustness in analyzing complex transcriptome variations.

Conclusions:

• MultiCluster offers a powerful approach for exploring multi-way interactions in large-scale gene expression datasets.
• Tensor-based methods provide advantages in incorporating information across similar tissues for enhanced gene discovery.
• This method advances the analysis of transcriptome complexity and biological variation.