Metagene projection strategies for discovering mechanisms of T cell differentiation in gene expression profiling data (58.13)

Abstract

Characterizing the phenotypic heterogeneity in memory T cells generated following antigen encounter is a major challenge in immunology. Tools such as CyTOF flow cytometry or gene expression profiling of T cells can capture increasingly large numbers of parameters. However analysis of high dimensional data with standard techniques such as principal component analysis or hierarchical clustering often fails to capture the biological basis for observed phenotypic heterogeneity. We address this problem by applying a metagene projection strategy based on independent component analysis and consensus non-negative matrix factorization clustering. This approach assumes that the gene expression profile of a population of CD8+ T cells is generated by groups of genes that are representative of different biological processes, and deconvolves the data into individual components (or metagenes) that together comprise some or all of the overall “signal” in the data-space. The metagene projection strategy is able to provide a robust low dimensional description of transcriptional (or other complex) data based on groups of genes that represent known and novel biological mechanisms. It is also able to provide sensitive clustering and classification solutions. We illustrate the ability of this metagene projection strategy to capture novel molecular patterns in T cell differentiation states common to antigen-specific CD8+ T cells responses to HIV, HCV and metastatic melanoma.