Data Analysis Approaches for High Content Screening

Abstract

Experimental methods in cell biology offer an excellent combination of access to detailed biologic information together with manageable cost. Recent technological advances have further enhanced this capacity, allowing the interrogation of cells in the automated high throughput mode that is necessary in the pharmaceutical industry. While the technology itself has advanced tremendously, accompanying statistical data analysis as often practiced has lagged behind. In order to address this issue, we embarked on a program whose goals are the identification of key aspects of data analysis and the development of appropriate statistical tools. We report several methods we have adopted that were found to be general and broadly applicable together with examples in which they were used. These include: the representation of cell-level information through quantiles of population distributions, capture of discrete processes and cell subpopulations by mixture modeling, construction of quantile scores and composite decision rules for analysis comprising multiple criteria, and robust multivariate supervised classification for hit selection with discovery of important features. No two studies are identical, however, and an array of statistical methods is needed, simple as well as complex, together with a flexible approach to carefully match the tools to the scientific questions at hand.

Keywords:

• High Content Analysis
• Quantile descriptors
• Random forest classifier

Acknowledgments

We thank Richard Raubertas, Ting Wang, and Vladimir Svetnik for their contribution to this work. We also thank the anonymous reviewers for their constructive comments that helped polish the article.