flowClean: Automated identification and removal of fluorescence anomalies in flow cytometry data

Abstract

Modern flow cytometry systems can be coupled to plate readers for high-throughput acquisition. These systems allow hundreds of samples to be analyzed in a single day. Quality control of the data remains challenging, however, and is further complicated when a large number of parameters is measured in an experiment. Our examination of 29.228 publicly available FCS files from laboratories worldwide indicates 13.7% have a fluorescence anomaly. In particular, fluorescence measurements for a sample over the collection time may not remain stable due to fluctuations in fluid dynamics; the impact of instabilities may differ between samples and among parameters. Therefore, we hypothesized that tracking cell populations (which represent a summary of all parameters) in centered log ratio (CLR) space would provide a sensitive and consistent method of quality control. Here, we present flowClean, an algorithm to track subset frequency changes within a sample during acquisition, and flag time periods with fluorescence perturbations leading to the emergence of false populations. Aberrant time periods are reported as a new parameter and added to a revised data file, allowing users to easily review and exclude those events from further analysis. We apply this method to proof-of-concept datasets and also to a subset of data from a recent vaccine trial. The algorithm flags events that are suspicious by visual inspection, as well as those showing more subtle effects that might not be consistently flagged by investigators reviewing the data manually, and out-performs the current state-of-the-art. flowClean is available as an R package on Bioconductor, as a module on the free-to-use GenePattern web server, and as a plugin for FlowJo X.

Computational flow cytometry: helping to make sense of high-dimensional immunology data

“…A first quality control step is to visualize scatter and marker values over time, and filter out regions that show abnormal behaviour (for example, due to clogging, speed change or air measurements when the tube is empty). Both the flowQ 34 and flowClean 35 packages provide this functionality. A second control step is to check for batch effects, which gives an idea of the between-sample variation.…”

International Society for Advancement of Cytometry (ISAC) flow cytometry shared resource laboratory (SRL) best practices

“…FlowClean, flowAI) (42,43) Provide QC results to users as needed to ensure good instrument operation during data acquisition Identify and communicate data sets that are erroneous. For example, using the time parameter in the FCS file or the inclusion of standard particles (e.g.…”

Flow cytometry data analysis: Recent tools and algorithms

“…flowAI evaluates three different properties: flow rate, signal acquisition, and dynamic range . flowClean detects anomalies in the data by tracking fluorescent measurement fluctuations within a sample during acquisition time …”