A key challenge in systems biology is to analyze emerging high-throughput image-based data to understand how cellular phenotypes are genetically encoded. We propose to study morphological variability as such a phenotype and develop a method to quantify morphological variation on a cell population level. Organisms have likely evolved to balance noise levels in signaling through regulatory networks in order to maximize phenotypic variability, without compromising the reliability of phenotypic responses. By applying our methods to data from a genetic screen, we identify new roles for genes as modulators of morphological variability consistent with their known biological function.
We apply our methods to a genetic screen consisting of 273 treatment conditions (TCs) where genes were systematically inhibited by RNAi, or overexpressed, in Drosophila migratory cells and described using 145 geometric features. The specific problem we address is the identification of genes that modulate noise levels present in different morphological processes. To do so in the context of data from a genetic screen requires a three-step process: first, defining a general measure of variability of single-cell populations; second, applying this measure to treatment conditions from the genetic screen; and third, identifying clusters of treatment conditions (and by extension, genes) which are implicated in control of a particular morphological process (e.g. adhesion formation). By comparing the variability measure across all treatment conditions within a given functionally-related cluster, we are able to identify genes that modulate morphological noise for a particular morphological process.
