Visualizing the structure of RNA-seq expression data using grade of membership models

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Grade of membership (GoM) models, also known as admixture or topic models, generalize cluster models by allowing samples to belong to multiple clusters.
• These models are established in population genetics for admixed individuals and in natural language processing for document topic analysis.

Purpose of the Study:

• To demonstrate the utility of GoM models for clustering RNA-sequencing (RNA-seq) gene expression data from bulk and single-cell samples.
• To develop and apply methods for interpreting these clusters by identifying distinctively expressed genes.

Main Methods:

• Application of GoM models to RNA-seq data.
• Development of methods to identify genes with distinct expression patterns within identified clusters.
• Utilizing the Bioconductor package CountClust for implementation.

Main Results:

• Successful clustering of RNA-seq data from the GTEx project (53 human tissues), highlighting tissue similarities and identifying key genes.
• Analysis of single-cell RNA-seq data from mouse preimplantation embryos, revealing developmental trajectories and associated gene functions.
• Demonstrated ability to identify both discrete and continuous variation in biological systems.

Conclusions:

• GoM models are effective for analyzing gene expression data, uncovering complex biological structures and heterogeneity.
• The developed methods provide valuable tools for interpreting expression patterns and understanding biological processes.
• CountClust package offers a robust implementation for these analyses in bioinformatics research.