Integrative statistical methods and tools for analysis of perturbation signatures

点击翻译按钮获取中文摘要

Revealing mechanism-based associations among effects of disparate biological perturbations by integrative analysis of diverse signatures requires statistical methods for identifying commonalities in highdimensional readouts of cellular states. It also requires statistical methods and computational algorithms for inferring and comparing condition dependent regulatory network models of perturbation effects. Generation and interpretation of transcriptional signature of diverse perturbations in diverse cell types is the focus one of the currently funded Large Scale Production of Perturbagen-induced Cellular Signatures (LINCS) centers (U54HG006093). The other cunrentiy funded center (U54HG006097) will assess signatures of signaling molecule activities and compound biochemical activity. Jointly, these diverse readouts of cellular states can be used to effectively elicit condition dependent regulatory network (CORN) models goveming the cellular response to perturbations. We propose to develop, test, validate, benchmark and implement a general statistical framework for assessing concordance in different types of perturbation signatures. Our preliminary results demonstrate stunning improvements in statistical power over currently available methods in identifying concordant transcriptional signatures. They also demonstrate the utility and strong improvements in statistical power for identifying common biological pathways by integrated analysis of signatures of multiple perturbations and multiple types of cellular readouts as opposed to using isolated analyses of different signatures. Based on these results we also propose integrative statistical methods for mechanistically explaining perturbation signatures in the context of known pathways and for constructing de-novo mechanistic network models. Finally, we propose a strategy for using newly developed methods and LINCS pertrubation signatures as a novel resource for intepreting disease-related genomics data. All new methods and algorithms will be deployed within existing on- and off-line computational platforms integrating data, computational tools and the functional knowledge base.