Imaging Noise and Bias Reduction for High-Content Screening

• Activity Number: 595
• Type: Invited
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Kerby Shedden*+
• Companies: University of Michigan
• Keywords: Cheminformatics ; Image analysis ; Measurement error

Abstract

High content screening is an experimental technique in which images of live cells are acquired following incubation with chemical agents. The technique is high-throughput, allowing thousands of compounds to be imaged in a single experiment. I will consider the estimation of numerical parameters from these images when the signal of interest is weak compared to the imaging noise, as occurs when imaging compounds that have not been optimized as bioimaging probes. Numerical parameters such as summaries of the cell-to-cell variation in subcellular distribution patterns become biased in this setting. The bias is not constant relative to the true value being estimated, and thus complicates the interpretation of many summary statistics. We propose a method based on simulation-extrapolation to substantially reduce this bias. As an illustrative example, we consider the degree of cell-to-cell variability at different points in the overall spectrum of subcellular localization behavior. Our analysis suggests that cellular heterogeneity is not constant along this spectrum. This is consistent with a hidden variation in the cells' biological states exerting an influence on their behavior.