Applications and characterization of mRNA expression compendia in inference of genetic association networks.

摘要

Large mRNA expression compendia, comprised of an organism's genome-wide responses to a range of experimental conditions, potently enable inference of genetic association networks. Here, we contribute to two applications of this strategy: perturbation target prediction and transcriptional regulatory network inference (TRNI).;A key challenge in analyzing an organism's expression response to a perturbation is distinguishing the genes directly affected from the many indirectly affected genes. To address this, we developed a genome-wide model of mRNA expression and applied sparse inference methods to construct a network model of gene interaction effects. This inferred network model was then used to filter the expression profile of a perturbation of interest and predict the genes directly targeted by that perturbation. Our method performed well in identifying targets of genetic perturbations in microarray compendia with up to a 32% improvement in sensitivity, compared to the next best method. The overall performance of our network-filtering method shows promise for identifying the direct targets of genetic dysregulation in cancer and disease from expression profiles.;In applying the method above, we observed evidence of structure within the compendia. We sought to explicitly characterize this structure and resulting dependency in an Escherichia coli microarray compendium ( n=376 experiments), and illustrate its consequences in performance and statistical testing for edge selection in TRNI. We observed structure across experimental condition-based groupings of experiments and, consequently, a substantially reduced effective sample size neff=14.7. Furthermore, we found that neff of select subsets of the data exceeded neff of the full compendium, and observed improved performance in TRNI using these subsets. Finally, using the RegulonDB truth-set, we demonstrated that false discovery rates derived using neff yielded accurate thresholds for edge selection in TRNI, while using n vastly overestimated the number of true edges. These results recommend neff as a potent, specific descriptor of microarray compendia and highlight a TRNI method that is readily applicable to compendia from any species, even when a truth-set is not available to guide edge selection.;These findings demonstrate improvement in perturbation target prediction and promote a more refined construction and utilization of microarray compendia in TRNI and related problems.