Skip to main content Skip to navigation

Computational Sorting With HyperFinder, FlowJo™ Software and BD FACSDiva™ Software



Gating novel cell populations

The use of machine learning tools like tSNE, UMAP, CITRUS and CellCNN for the analysis of flow cytometry data may result in the discovery of novel cell types with a distinct pattern of surface markers. However, the novel cell population cannot be sorted for further investigation without a gating strategy, as current sorters require a series of 1D or 2D gates for sort decisions to be made.


In principle, the solution to this problem can be obtained by creating all possible gating strategies and determining which one best captures the target population. This goodness of fit of a gating strategy can be quantified through yield (the percentage of target cells that are collected in the last gate) and purity (the percentage of target cells in the last gate that belong to the population of interest). For the sake of simplicity, those metrics can be combined into a simple F score, the harmonic mean of yield and purity. Therefore, the correct gating strategy for a target population would be the one with the highest F score for that population.



Finding the right gating strategy

Algorithms like GateFinder and Hypergate have been proposed in the past, that help to solve the problem of finding the right gating without having to screen the large number of possible gating strategies one by one. However, both approaches have their limitations. GateFinder, which selects the right gate from a set of candidate gates created by analysing all biaxial plots, drawing a convex gate around the population of interest and iterating from a sequence of convex gates, is effective only on low-dimensional datasets. In contrast, Hypergate, which fits a multi-dimensional box (hyper-rectangle) around the population of interest and creates a set of rectangular gates for the search algorithm, often fails when real-world cases are not captured by the rectangular gate. Both approaches require the knowledge of the R programming language, reducing their ease of use, and also do not output the gates in a format that can be read by FACS software like BD FACSDiva™ Software.



HyperFinder workflow

HyperFinder combines the core principles of GateFinder and Hypergate to optimise gating strategies. It takes an input of cells labelled as positive and negative and tries to find a gating strategy that captures the maximum number of positives (maximising yield) while removing the negatives (maximising purity). HyperFinder balances both parameters by using an Fβ score, which is a weighted harmonic mean of yield and purity in which the β parameter allows the user to bias the selection strategy in favour of either of the two.


HyperFinder is implemented as a FlowJo™ Software plugin with a simple interface that does not need any specialised programming knowledge. With FlowJo™ v10.6 Software or higher, users can export HyperFinder-generated gates directly into a BD FACSDiva™ Software worksheet for use on the BD FACSAria™ Cell Sorters or the BD FACSymphony™ S6 Cell Sorter for a complete clustering analysis-to-sorting workflow.


This white paper discusses the use of HyperFinder for sorting cells utilising the BD FACSDiva™ Software in three test cases. All FlowJo™ Software plugins are available at FlowJo Exchange.


Read the paper: Computational Sorting With HyperFinder, FlowJo™ Software and BD FACSDiva™ Software.



Purity Field Yield



BD, the BD Logo,  FACSAria™, FACSDiva™, FACSymphony™ and FlowJo™ are trademarks of Becton, Dickinson and Company or its affiliates. © 2022 BD. All rights reserved.

Products are for Research Use Only. Not for use in diagnostic or therapeutic procedures.

BD FACSAria™ and BD FACSymphony™ S6 are Class 1 Laser Products