Prediction and expansion of biological pathways from perturbation effects

Complex phenotypes, such as cancer invasion, result from the actions and interactions of many genes and gene products. Though pathway-based analysis can offer improved predictions over single-gene or set-of-gene analyses, few pathways have been characterized. New high-throughput technology offers the opportunity for individual investigators to learn entire pathways from a small number of gene perturbation and gene expression experiments. I present two pathway inference methods using data from downstream perturbation effects and use the inferred structures to predict novel pathway members in cancer invasion and Vibrio cholerae biofilm.;In a V. cholerae system, microarray gene expression data under gene perturbations from deletion knockouts was analyzed using a new method called a Joint Intervention Network. This analysis resulted in an inferred regulatory network of the perturbed genes, and prediction of biofilm-associated genes that was more accurate than a correlation-based method.;I next developed a signed version of the Nested Effects Model and an associated efficient structure inference method, named Factor Graph-Nested Effects Model (FG-NEM). On synthetic data I show improved performance of FG-NEM over unsigned versions of the algorithm. In yeast, FG-NEM predicts Gene Ontology categories more accurately than a correlation-based method. And finally, in a cancer cell line I predicted an invasion network and identified fourteen new genes necessary for cancer invasion.