Automation of flow cytometry data analysis with elastic image registration

Abstract

Cell populations in flow cytometry are typically identified via visual manual gating, a time-consuming and error-prone approach to select subpopulations based on expression of cellular markers. Batch processing can be used to automate the analysis of bimodally distributed data but underperforms with highly-variable or continuously-expressed markers. We developed a visual pattern recognition automated gating tool, BD ElastiGate Software (hereafter ElastiGate), to recapitulate the visual process of manual gating by automatically adjusting gates to capture local variability. ElastiGate converts histograms and two-dimensional plots into images, then uses elastic B-spline image registration to transform pre-gated training plot images and their gates to corresponding ungated target plot images, thereby adjusting for local variations. ElastiGate was validated with biologically relevant datasets in CAR-T cell manufacturing, tumor-infiltrating immunophenotyping, cytotoxicity assays (> 500 data files), and a high-dimensional dataset. Accuracy was evaluated against corresponding manually gated analysis using F1 score statistics. ElastiGate performed similarly to manual gating, with average F1 scores of > 0.9 across all gates. ElastiGate, accessible as a FlowJo Software plugin or in BD FACSuite Software, uses minimal training samples to automate gating while substantially reducing analysis time and outperforming existing 2D plot autogating solutions in F1 scores and ease of implementation.

Fig. 1

ElastiGate application on three benchmarking datasets. (A) The distributions of F1 scores for each gate over 30 target fcs files in the lysed whole blood scatter dataset, comparing ElastiGate to gates drawn by three manual expert analysts (Manual 1–3). Lymphocytes, Granulocytes, and Monocytes gates on FSC-A versus SSC-A plot were provided from one training file. The heatmap bar represents number of cells in each gate as a Log10 scale. Values represent the mean and error bars represent standard error of the mean (SEM). (B) The number of cells in each population plotted against its ElastiGate F1 score from the lysed whole blood scatter dataset. (C) The distributions of F1 scores for each gate over 30 target fcs files in the multilevel fluorescence quantitation beads dataset. (D) The distributions of F1 scores for each gate over 20 fcs files in the monocyte subset analysis dataset. ElastiGate was used with one training file.

Fig. 2

ElastiGate application on a CAR-T cell manufacturing dataset. ElastiGate was used with 3 training files for all apheresis target fcs files and 1 training file for all final product fcs files. The first 5 cleanup gates were applied to all files, and the remaining gates were applied to non-Isotype fcs files. (A) The distributions of F1 scores for the first 5 gates over all 71 apheresis and final product target fcs files comparing ElastiGate to gates drawn by three manual expert analysts (Manual 1–3). Values represent the mean and error bars represent standard error of the mean (SEM). The heatmap bar represents number of cells in each gate as a Log10 scale. (B) The distributions of F1 scores for the last 7 gates, for 35 apheresis target fcs files (Isotype files are excluded). (C) The distributions of F1 scores for 3 gates, for 11 final product target fcs files (Isotype files are excluded). The % Transduction gate was calculated using the “Sample” final product fcs files. Other gates were excluded due to a small cell number, with < 30 cells. WBC: White blood cells.

Fig. 3

ElastiGate application on a tumor-infiltrating lymphocyte immunophenotyping dataset. (A) The mean F1 score for all gates in the tumor-infiltrating lymphocyte immunophenotyping dataset, using 2, 4 and 6 training files. Statistical analyses were performed with non-parametric paired One-Way Anova and the p values were corrected for multiple comparison (Friedman test), *p < 0.05. (B) The distributions of F1 scores for each gate over 40 target fcs files in a tumor-infiltrating lymphocyte immunophenotyping dataset comparing ElastiGate to gates drawn by four manual expert analysts (Manual 1–4). Values represent the mean and error bars represent standard error of the mean (SEM). ElastiGate was used with 6 training files. The heatmap bar represents number of cells in each gate as a Log10 scale.

Fig. 4

ElastiGate in a high throughput cytotoxicity assay. (A) The distributions of F1 scores for each gate over 425 target fcs files in the high throughput cytotoxicity assay comparing ElastiGate to gates drawn by six manual expert analysts (Manual 1–6). ElastiGate was used with 15 training files. Values represent the mean and error bars represent standard error of the mean (SEM). The heatmap bar represents number of cells in each gate as a Log10 scale. (B) F1 scores of ElastiGate versus manual expert analyst (orange), FlowDensity versus manual expert analyst (blue), and Cytobank Automated gating versus manual expert (black) from a 38-file subset (one plate) of the dataset. Error bars represent standard error of the mean (SEM). CTV: CellTraceViolet.

Fig. 5

ElastiGate application on 38-color T-cell phenotyping panel. (A) The distribution of mean F1 scores comparing ElastiGate to 3 manual analysts for each gate (49 gates in total), with the error bars representing standard error of the mean (SEM). ElastiGate was run on 29 target files using 3 files for training. The heatmap bar represents number of cells in each gate as a Log10 scale. (B) The number of cells in each population plotted against it’s ElastiGate F1 score from the 38-color T-cell phenotyping panel.