UC San Diego

Community detection in complex networks (graphs) has been the subject of investigation in numerous domains. In biological networks, communities are functionally contextual and often provide insights into mechanisms. Detecting communities and analyzing their biological functions is an important aspect of studying biological networks. Communities (aka modules) can yield useful insights into the structure of networks and serve as a basis for analyzing them at topological and functional levels. The work presented in this dissertation is aimed at community detection in human disease (specifically colorectal cancer) networks using different approaches, with the focus on analyzing their biological functions.The study begins with the exploration of existing community detection algorithms and evaluation of their findings on two important Protein-Protein Interaction (PPI) networks, namely, Saccharomyces cerevisiae (Yeast) and Homo sapiens (Human) at both topological and functional levels. The main criteria to assess the performance of each method are 1) appropriate community size (neither too small nor too large), 2) representation of only one or two broad biological functions within a community, 3) most genes from the network belonging to a pathway should also belong to only one or two communities, and 4) performance speed. These criteria enable us to select one of the best methods for detecting communities in biological networks. Next, a gene expression microarray dataset of colorectal cancer (CRC) is analyzed to detect stage-specific biomarkers as well as modular mechanisms, potentially causal for the progression of CRC from normal to stages I to IV. Constructing unweighted and weighted correlation networks for each stage, communities are identified and compared topologically and functionally across stages. Short Time-series Expression Miner (STEM) algorithm is also used to detect potential biomarkers having a role in CRC. Constructing a drug-target-PPI network provides insight, in the light of analyzed data, into understanding the functional mechanisms for some of the current drugs used in CRC treatment. Lastly, gene modules across stages of CRC are analyzed from a single cell transcriptomic dataset to decipher mechanistic changes likely contributing to tumor growth and cancer progression. Cell down-sampling process is firstly performed at stages pT2 to pT4 (as well as right colon) to make cell count equal across different stages and tissue sites. Functional modules at early stage (pT1) and also right colon are identified utilizing Weighted Gene Co- expression Network Analysis (WGCNA). In particular, WGCNA’s preservation statistics are used to detect gene modules that exhibit weak/strong preservation of network topology in late stages (pT234) vs. early stage (pT1) as well as left colon vs. right colon. Functional enrichment analysis of the non-preserved modules reveals mechanisms related to the initiation, progression and metastasis of CRC.