Personalization of graphical models using multi-omics data for subtype discovery and prognosis - SUMMARY
Recent clinical advances in cancer treatments have been attributed to targeting specific genes such as ER-,
HER2, etc. However, a significant percentage of patients do not respond to targeted therapies or develop
resistance over time. This implies that current methods for tumor characterization and therapeutic interventions
are not sufficiently accurate. In particular, current disease/patient subtyping approaches all look for differences
at the level of individual genes, ignoring pathway-level interactions that can hold key characteristics of cancer
disparities. The main goal of this project is to pioneer a new approach to disease/patient subtyping that
departs from the traditional paradigm: subtyping and characterization at the pathway level, using personalized
pathway profiles, rather than at the gene level. The hypothesis driving this work is that an emerging condition
for an individual patient can be triggered through different genes and molecular levels (e.g., transcriptome,
epigenome, etc.) but might involve the same mechanism(s). This is because, while alterations of impacted
genes could be very diverse between patients the pathways involved could be the same. The innovation of this
work is the development of a novel approach able to compute pathway profiles of individual patients by
effectively taking into account gene topology and pathway crosstalk. Fundamental to this approach is effective
integration of multi-omics and multi-cohort data to take advantage of complimentary information among
different data types and address the small size problem associated with many cohorts. The goal of this project
will be achieved through four specific aims: 1a) identify impacted pathways in individual patients, 1b) integrate
mutation, copy number variation, methylation, microRNA, and gene expression, 2) integrate multi-cohort data,
3) identify pathway signatures for each subtype, and 4) validate the proposed pathway-level subtyping
methodology and associated risk prediction by leveraging public data as well as data from two clinical studies
at UPMC Hillman Cancer Center. The significance of the proposed work lies on its potential to provide new
methods and tools for better cancer management and prognosis. In the longer term, personalized pathway
analysis will improve our understanding of disease mechanisms and resistance to treatments, enabling the
development of new treatments for personalized medicine. The methods and tools will be made available
through an open-access web application and a CRAN R package.
