Evaluation of A Novel — Distribution Sensitive — HCS Summary Statistic
1. Department of Human Genetics, McGill University, Montreal, Canada; 2. McGill University and Genome Quebec Innovation Centre, Montreal, Canada
The analysis of high-content screening (HCS) assays routinely aggregates cell measurements within wells into a single measure, thereby reducing resolution and discarding potentially useful information regarding cell subpopulations. For this reason, a designed HCS assay with known compounds and effects could show potential advantages when cell population distributions are compared between wells. An examination is done on the morphology of cells treated with two compounds with well-known effects on cells and in particular, on the fibrous microtubule structure, in a data set of thousands of confocal microscope images of HeLa cells. The performances of 47 cell metrics (aggregated or individually assessed) to distinguish between active and inactive compounds will be compared using lysed and non-lysed cells in raw images versus background corrected images. First an assessment is carried out of potential improvements in measurement accuracy and precision obtained with a recently developed method (IQEM) that corrects for non-uniformity of background light intensity and is specific to HCS images. Second a comparison of individual measurements (mean/median) to aggregate cellular measurements is performed. Aggregate measurements are extracted from ranking of treated cells against controls to plot ROC curves and then calculate area under the curve (AUC) values. Finally various cell metrics are compared, in particular Fiberscore metrics which are specifically designed to measure cellular fiber networks and which have not been previously used within HCS context. The motivating biological problem is examination of the effects of two types of drugs on the microtubule structure of cytoskeleton with implications for understanding fundamental biology and for drug development.