Annotating Human Interactome to Predict Pathways and Systematically Analyzing Network Rewiring in Cancer across Multiple Tissues

Human protein-protein interaction (PPI) networks are essential in regulating most cellular processes and their disruption can cause complex diseases such as cancer. Currently, hundreds of thousands of binary PPIs comprise human interactome, but they lack annotations about tissue- or disease-specificity. Pathway databases provide more detail about these conditions, but they cover only fraction of the proteome. The main goal of this dissertation was to annotate interactome through integrative computational methods and use these annotations, to predict novel protein-pathway associations (Chapter 2), and perform system-level analysis of interactome in cancer (Chapter 3). Chapter 2 describes integration of 4,968 pathways from twenty curated pathway databases. Analysis of these data highlighted several challenges including low proteome coverage (63%), low overlap across different databases, unavailability of direct information about underlying physical connectivity of pathway members, and high fraction of protein-coding genes without any pathway annotations, i.e., 'pathway orphans'. To address these challenges, I integrated these data with PPIs to predict biologically relevant protein-pathway associations. Cross-validation determined 71% recovery rate of our predictions. Data integration and predictions increase coverage of pathway annotations for proteins to 91% and provide novel annotations for 5,402 pathway orphans. These data are publicly available in pathDIP (http://ophid.utoronto.ca/pathDIP). Chapter 3 describes curation of gene expression data of 3,127 primary cancer and normal samples across sixteen tissues and thirty-five cancer types. I used pairwise gene coexpression and differential coexpression to annotate 276,552 PPIs with tissue- and cancer-specificity. Pan-cancer comparison of annotated PPIs provides novel insights toward dynamics of PPI networks in tissue-cancers. We found that while PPI network hubs are present in most of the tissues and cancers, their interactions are highly tissue-cancer-specific. Comprehensive graph theory-based analysis shows that highly dense network modules (e.g., protein complexes) are frequently maintained in both normal and cancer, but they are wired differently. I demonstrate on two examples of the potential of these annotations in discovering novel cancer genes and drug pathways.

Department dc:contributor.department

Medical Biophysics

Year dc:date.issued

2018

Author dc:creator

• Rahmati, Sara

Advisor dc:contributor.advisor

• Jurisica, Igor

• biological networks analysis
• computational and integrative biology
• data integration
• high-throuput data analysis
• molecular pathways
• protein-protein interactions

Handle dc:identifier.uri

http://hdl.handle.net/1807/91994

OAI identifier oai:identifier

oai:utoronto.scholaris.ca:1807/91994