Attractor Metafeatures and Their Application in Biomolecular Data Analysis

This dissertation proposes a family of algorithms for deriving signatures of mutually associated features, to which we refer as attractor metafeatures, or simply attractors. Specifically, we present multi-cancer attractor derivation algorithms, identifying correlated features in signatures from multiple biological data sets in one analysis, as well as the groups of samples or cells that exclusively express these signatures. Our results demonstrate that these signatures can be used, in proper combinations, as biomarkers that predict a patient’s survival rate, based on the transcriptome of the tumor sample. They can also be used as features to analyze the composition of the tumor. Through analyzing large data sets of 18 cancer types and three high-throughput platforms from The Cancer Genome Atlas (TCGA) PanCanAtlas Project and multiple single-cell RNA-seq data sets, we identified novel cancer attractor signatures and elucidated the identity of the cells that express these signatures. Using these signatures, we developed a prognostic biomarker for breast cancer called the Breast Cancer Attractor Metagenes (BCAM) biomarker as well as a software platform to analyze the tumor sample, called Analysis of the Single-Cell Omics for Tumor (ASCOT).