Explicit Representation of Protein Activity States Significantly Improves Causal Discovery of Protein Phosphorylation Networks (Jinling) and A Radiogenomics Approach to Predicting Immune and Stromal Cell Line Invasion in Breast Cancer Lesions (Ryan)

Seminar Date: 
Seminar Time: 
11am - 12pm
Seminar Location: 
5607 Baum Boulevard, Room 407A
Jinling Liu & Ryan Hausler

Jinling Liu:

Protein phosphorylation networks play an important role in cell signaling. In these networks, phosphorylation of a protein kinase usually leads to its activation, and the activated kinase will in turn phosphorylate downstream target proteins.  These phosphorylation networks encode causal relationships between proteins, which can be inferred through causal inference algorithms, such as the Fast Greedy Equivalence Search (FGES). Prior efforts have applied causal inference algorithms to phosphoprotein abundances data, assuming that the phosphorylation status of proteins (reflected by phosphoprotein abundances) is indicative of protein activity states. However, the phosphorylation status of a kinase does not necessarily reflect its activity state, especially when interventions such as inhibitors or mutations directly affect its activity state without changing its phosphorylation status. We developed a causal discovery framework which explicitly represents the activity state of each protein kinase as an unmeasured latent variable, and we developed a novel algorithm to infer the protein activity state by integrating information from measured phosphoprotein abundances, prior knowledge of protein phosphorylation networks, and pharmacological interventions. We further applied the FGES algorithm to search for causal relationships between the activity state of a kinase and the phosphorylation status of its target proteins.  This framework allows us to effectively represent the impact of interventions on protein activity states as well as to model the stochastic relationships between phosphoprotein abundances and activity states of kinases.  We applied our framework to a reverse phase proteomic array dataset derived from breast cancer cell lines. The results showed that our approach significantly enhanced the capability of FGES in searching for the true protein phosphorylation networks.


Ryan Hausler:

Studies have shown that prognostic outcomes of tumors are not only linked with genetic and epigenetic factors within the cancerous cells, but also with the extent of infiltrating immune and stromal cells in the tumor microenvironment (TME). In clinical practice, evaluating the TME using traditional radiology has not proven effective, leaving pathology as the best option to measure the presence of immune and stromal cells. Radiomics, however, has shown the potential to evaluate the TME, possibly leading to radiological predictors of a cancer patient’s response to treatment and a better understanding of how immune and stromal cell infiltration can affect a tumor’s visible phenotype. In this study, we create machine learning models utilizing DCE-MRI features that effectively predict the presence of eight immune and two stromal cell populations within breast cancer lesions.