Title: "A Network Topology Approach to Selecting Proteins for Single Cell Studies of Signalling Cascades"
Speaker: Robert Ness*; Department of Statistics, Purdue University

Place: Materials and Electrical Engineering (MSEE) B012
Date: April 14, 2015; Tuesday
Time: 4:30pm


Cells sense and respond to diverse stimuli through signal transduction. In this process, an input signal, (e.g., ligand-receptor binding), triggers a cascade of phosphorylation reactions inside the cell, leading to a signalling response (e.g., transcription of a gene). The set of proteins and reactions in this cascade constitute a signalling network. In the study of signalling networks, cellular heterogeneity has motivated techniques that allow for simultaneous measurement of multiple signalling proteins inside a single cell. The cutting edge of such technologies is mass cytometry, which uses mass spectrometry to quantify the abundance of of signalling proteins in individual cells.

By applying Bayesian network structure learning algorithms to single cell measurements, we can can infer the signalling network structure. The goal of inference is to discover novel signalling events specific to the experimental context. However, structure learning algorithms rely on faithful measurement of all key proteins on the signalling pathways active in the measured cells. As the number of proteins that can be measured by mass cytometry is bounded (~50), one must prioritize proteins by how informative they are likely to be of signalling dynamics and ultimately cellular phenotype.

We propose a method that prioritizes proteins for measurement based on signalling network topology. The approach has two inputs; a canonical signalling pathway map (e.g., KEGG pathway), and prior single cell data on some subset of the proteins on the pathway. From these inputs we fit a computational model of the signalling system, with parameters corresponding to kinetic rate laws.

Results: We apply the approach to simulated data, as well as actual single cell measurements. The approach prioritizes proteins by identifying areas of the computational model with greater uncertainty. We evaluate results by ability to recover novel signalling events hidden from the input topology.

*Collaborators: Parag Mallick1; Olga Vitek2
1Stanford University, Palo Alto, CA; 2Northeastern University, Boston, MA

Click here www.stat.purdue.edu/~doerge/BIOINFORM.D/SPRING15/sem.html for a full schedule of BIOINFORMATICS SEMINARS, past and present.