Image-based cell profiling enhancement via data cleaning methods
Open Access
- 4 May 2022
- journal article
- research article
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 17 (5) , e0267280
Abstract
With the advent of high-throughput assays, a large number of biological experiments can be carried out. Image-based assays are among the most accessible and inexpensive technologies for this purpose. Indeed, these assays have proved to be effective in characterizing unknown functions of genes and small molecules. Image analysis pipelines have a pivotal role in translating raw images that are captured in such assays into useful and compact representation, also known as measurements. CellProfiler is a popular and commonly used tool for this purpose through providing readily available modules for the cell/nuclei segmentation, and making various measurements, or features, for each cell/nuclei. Single cell features are then aggregated for each treatment replica to form treatment “profiles”. However, there may be several sources of error in the CellProfiler quantification pipeline that affects the downstream analysis that is performed on the profiles. In this work, we examined various preprocessing approaches to improve the profiles. We consider the identification of drug mechanisms of action as the downstream task to evaluate such preprocessing approaches. Our enhancement steps mainly consist of data cleaning, cell level outlier detection, toxic drug detection, and regressing out the cell area from all other features, as many of them are widely affected by the cell area. Our experiments indicate that by performing these time-efficient preprocessing steps, image-based profiles can preserve more meaningful information compared to raw profiles. In the end, we also suggest possible avenues for future research.Keywords
This publication has 20 references indexed in Scilit:
- Comparison of Methods for Image-Based Profiling of Cellular Morphological Responses to Small-Molecule TreatmentSLAS Discovery, 2013
- An approach for extensibly profiling the molecular states of cellular subpopulationsNature Methods, 2009
- CellClassifier: supervised learning of cellular phenotypesBioinformatics, 2009
- Anomaly detectionACM Computing Surveys, 2009
- Statistical practice in high-throughput screening data analysisNature Biotechnology, 2006
- Multidimensional Drug Profiling By Automated MicroscopyScience, 2004
- A brain tumor segmentation framework based on outlier detection*1Medical Image Analysis, 2004
- Discovering cluster-based local outliersPattern Recognition Letters, 2003
- LOFACM SIGMOD Record, 2000
- Learning representations by back-propagating errorsNature, 1986