A0997
Title: Automated statistical methods for high-throughput phenotyping experiments
Authors: Elif Acar - University of Manitoba (Canada) [presenting]
Abstract: Many health applications produce ever-increasing quantities of biological data. As such applications often rely on automated pipelines for data analysis, an important statistical challenge is to evaluate and refine these pipelines as more and more data are acquired. This challenge is exemplified by the high-throughput phenotyping experiments conducted by the International Mouse Phenotyping Consortium (IMPC), where multiple phenotype measurements are obtained for a small set of gene-edited mice and a large set of controls acquired continually over time. Model selection is a fundamental component of the automated pipeline, increasing the power of detecting the gene effect. However, the effect of post-selection inference in this setting is not well understood. Moreover, due to the size and complexity of the data, gene function is assessed by combining the results of univariate phenotype analyses. However, analyzing multiple phenotypes simultaneously at the individual level greatly improves the power of detection. The focus is on evaluating and improving the IMPC statistical pipeline along these lines of inquiry.
