FlowJoBridge

An interoperability framework between MATLAB and FlowJo.

This submission provides read/write access to data and population (gate) definitions stored in the workspaces of FlowJo 10.x. FlowJo is the most widely used software for single cell analysis, which is a cornerstone not only of immunology and vaccine research, but also of fields like oceanography and plant research. This submission primarily aims to foster the development of automatic cell population classifiers so as to deprecate the widely adopted practice of manual gating, which is time consuming and frequently subjective.

The function SupervisedGatingOn2MiceStrains in the file fcs/FlowJoBridgeExamples.m illustrates

• Training MATLAB’s supervised classification tools with a FlowJo workspace’s peer reviewed manual gates.
• Updating the FlowJo workspace with the gates that this training predicts on other samples

function SupervisedGatingOn2MiceStrains

method=questdlg('Do MATLAB supervised gating with ...',...

'Which method?', 'Discriminant analysis', ...

'Neural network', 'Discriminant analysis');

gating='/Singlets/Live singlets';

workspace=['https://storage.googleapis.com/' ...

'cytogenie.org/GHOSN2.wsp'];

%TRAIN fitcdiscr or fitcnet using the data and 9 parameters

% used to manually gate a representative sample in FlowJo

columns={'FSC-A', 'SSC-A', 'CD5:*', 'CD11b:*', 'F4/80:*', ...

'CD11c-biot:*', 'CD19:*', 'IgD:*', 'IgM:*'};

flowJoURI=['all_3-3.fcs' ... %BALB/c mouse strain

gating '@' workspace];

[data, ~, labels]=FlowJoTree.Read(flowJoURI, columns, 'labels');

if isequal(method, 'Neural network')

training=fitcnet(data, labels, 'Verbose', true, ...

'VerboseFrequency', 25, 'IterationLimit', 100,...

'LayerSizes', [100 50 25], 'Standardize', 1);

else

training = fitcdiscr(data, labels);

end

%NOW use this training to predict populations on a sample from

%a different mouse strain stained by different reagents

flowJoURI=['all_3-4.fcs', ... %genetic knockout strain

gating '@' workspace];

[data, ~, trueLabels,~, ~,~,~,~, gates]...

=FlowJoTree.Read(flowJoURI, columns, 'labels');

predictedLabels=predict(training, data);

%visualize the predictions with MATLAB’s confusion chart

confusionchart(figure, trueLabels, predictedLabels, ...

'Title', {[method ' prediction of cell ' ...

'populations in a RAG mouse'], ...

['strain (lacks B & T cells) from training on a ' ...

'BALB/c mouse strain']});

%write back the predictions into the FlowJo workspace

FlowJoTree.CreateLabelGates([method ' gating'],...

'demoIcon.gif', data, predictedLabels, [], columns, gates);

end

The example calls to fitcnet and fitcdiscr have many other options including auto tuning that the program can explore for each separate dataset.

The submission also provides

• Much additional functionality that we have developed for the analysis of flow cytometry data.
• An intuitive GUI that enables the non-programmer to benefit immediately from the power of MATLAB when evaluating and extending FlowJo’s gating definitions.
• Access to 9 demo workspaces that access the data and gating definitions of recent publications.

To start the GUI and set the MATLAB paths, invoke the “fjb” command from the root folder where you unzip our submission.

This submission complements the functionality of FlowJo, but does not replace it. Data setup and the majority of manual gating must be done in FlowJo and gate definitions can only be adjusted or created in FlowJoBridge for pre-existing X/Y parameter pairings.

This submission is provided by the Herzenberg Lab at Stanford University.

FlowJo is a registered trademark of FlowJo, LLC, a wholly owned subsidiary of Becton, Dickinson and Company (BD). Visit www.flowjo.com for more information about FlowJo. This File Exchange submission is not affiliated with FlowJo, LLC or BD.