Noise Reduction in Genome-Wide Perturbation Screens Using Linear Mixed-Effect Models

• Activity Number: 464
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Danni Yu*+ and John Danku and Ivan Baxter and Sungjin Kim and Olena K. Vatamaniuk and David E. Salt and Olga Vitek
• Companies: Purdue University and Purdue University and U.S. Department of Agriculture and Cornell University and Cornell University and Purdue University and Purdue University
• Keywords: statistical analysis ; linear mixed-effect models ; noise reduction ; genome-wide perturbations ; high-througput screening ; data normalization

Abstract

High-throughput perturbation screens measure the phenotypes of thousands of biological samples under various conditions, which are subject to substantial biological and technical variation. Meanwhile, it is often impossible to include a large number of replicates and to randomize the order of the replicates in high throughput screens. Distinguishing true changes in the phenotype from stochastic variation in such experimental designs is extremely challenging, and requires adequate statistical methodology. We propose a statistical modeling framework that is based on experimental designs with at least two controls profiled in the experiments, and a normalization and variance estimation procedure with linear mixed-effects models. We evaluate the framework using three comprehensive screens of yeast: 4940 gene-deletion haploid mutants, 1127 gene-deletion diploid mutants, and 5798 gene-overexpression haploid mutants. The proposed approach can be used in conjunction with practical experimental designs, allows extensions to alternative experimental workflows, enables a sensitive discovery of biologically meaningful changes, and strongly outperforms the existing noise reduction procedures.