Insights from Systematically Analyzing Microbial Phenotypic Profiles

Abstract

Following classical genetic approaches to understanding gene function, high-throughput phenotyping methods have emerged as a new way of studying gene functions, especially in microorganisms, which are highly amenable to high-throughput experimental design. As more high-throughput microbial phenotype data as well as the low-throughput data become available, systematically managing, displaying, and analyzing these data become a pivotal part in discovering unknown functions for genes. In this work, I have curated some datasets for high-throughput microbial phenotype data that contain genomic-scale phenotypes from E. coli tested under hundreds of conditions. Next, I conducted systematic and unbiased statistical analysis of these phenotype datasets and showed that the phenotypic profiles within these datasets are highly correlated with various functional annotations. The phenotype-function correlation has also been seen when a curated cell-cycle related phenotypic profile of S. cerevisiae is used with Gene Ontology annotations. Furthermore, I have displayed the preliminary results of using machine learning techniques to predict gene functions using high-throughput phenotype data of complete annotations, given more functional annotations as labels. Lastly, I describe a software package written in R that is potentially useful in analyzing high-throughput microbial phenotype data.