Skip to main content
eScholarship
Open Access Publications from the University of California

Sample-Specific Cancer Pathway Prediction From Genomic, Transcriptomic and Phosphoproteomic Data

  • Author(s): Paull, Evan Oliver
  • Advisor(s): Stuart, Joshua M
  • et al.
Creative Commons 'BY-SA' version 4.0 license
Abstract

Cancer phenotypes such as invasion, evasion of programmed cell death, and rapid growth, arise from the complex interactions of genes, proteins and extra-cellular environments. Understanding how selective alterations in the genome convert healthy cells to cancer is a critical step in developing new targeted and combination therapies. Current technology allows for detailed measurement of both genomic state as well as phenotype, through measurement of gene expression, chromatin state and protein activation. I present a method, Tied Diffusion through Interacting Events (TieDIE), that uses a “heat diffusion” model of information transfer to find pathways linking key genomic alterations to phenotypic effects, using high-throughput data collected from cohorts of cancer patients. Applying this method to four large data sets developed by The Cancer Genome Atlas (TCGA), I found key genes and interactions linking mutations related to histone modification and protein kinase signaling to gene-expression signatures of growth and proliferation. In a TCGA study of thyroid carcinoma, TieDIE found key proteins that modulate oncogenic signaling from mutant BRAF and RAS proteins to downstream MEK and ERK pathways, including the kinase suppressor of ras 1 scaffold protein.

TieDIE was next applied to a study of prostate cancer that included detailed measurements of the phosphpoproteome. With collaborators at UCLA, we used tissue from lethal, metastatic castration-resistant prostate cancer (CRPC) patients obtained from rapid autopsy to evaluate diverse genomic, transcriptomic, and phosphoproteomic datasets for pathway analysis. Using TieDIE, I integrated transcriptional, genomic and phosphoproteomics datasets to reveal a “map” of activated kinases and signaling pathways in CRPC. In contrast to single-dataset analyses, I show this integrative approach provides a more comprehensive and detailed look at metastatic signaling, and is generally useful in combining diverse datasets with only partially overlapping samples. Patient-specific network models were created by intersecting each sample’s data and protein activity predictions with the CRPC signaling “map.” These models reveal a hierarchy of the top kinase targets for each patient analyzed, and the corresponding therapeutic intervention, allowing for the construction of feasible strategies for patients with high activities in multiple kinases.

Main Content
Current View