Jump to main content.

2007 CompTox Forum

Abstract - Combined Data-Driven Biomedical Outcome Prediction and Interaction Network Inference from Molecular Profiling Data

Back to Agenda

Roland Somogyi, Ph.D.
President
Biosystemix, Ltd.
1090 Cliffside La. RR1PO
Sydenham, Ontario, K0H2T0, Canada
Phone: 613-376-3126
E-mail: rsomogyi@biosystemix.com

Large-scale molecular profiling of cells and tissues in response to toxicological perturbations provides valuable information that can be used for two key purposes: (1) sensitive and reliable monitoring of toxicological outcomes based on predictive molecular activity signatures; and, (2) discovery of the molecular interaction networks that underlie toxic responses. Using a new fundamental computational method, Predictive Interaction Analysis (PIA), we can infer synergistic and competitive relationships from pair-wise interactions of activity variables in distinguishing key biomedical tissue/cell and response classes. Such variable pairs generally perform with better outcome separation accuracies and statistical p-values compared to single variables. Therefore, pairs of molecular activity variables represent a valuable approach to better measuring the degree and type of toxic outcomes from in vivo and in vitro molecular profiling tests. From a systems biology perspective, the comprehensive list of PIA activity variable pairs represents a connectivity matrix, which can be displayed in the form of a molecular interaction network graph. Using this data-driven pathway discovery approach, we can visualize and study networks of synergistic and competitive molecular interaction relationships. These network models provide access to valuable insights into the nature and control points of toxic response, as directly derived from the data using the straightforward, statistically conservative PIA methodology.

Local Navigation

Jump to main content.