Version 1, November 2021

 

This demo teaches how to compare AutoGate’s fully automatic gating to the cell types defined in OMIP’s published gating strategy - http://cgworkspace.cytogenie.org/Tutorials/omip44.png .  

For reference, full publication is available here - https://onlinelibrary.wiley.com/doi/10.1002/cyto.a.23331

 

This document describes how to start in any position of a conventional gating hierarchy and run unsupervised automatic gating (EPP) and then compare the results between it and any other gating method AutoGate’s HiD matching tools and highlighting tools.

1      Set up the experiment

1.1       Click “Open AutoGate demo experiment” on the main window

AutoGate responds by showing you the below list of demo experiments available as of November 2021

 

1.2       Pick OMIP-44: 28 Color Immunophenotyping Human Dendritic

 

AutoGate responds by opening up the GatingTree window on the publication’s 4 full stained samples.  The bundled demo has already completed the setup plus the replication of the published manual gating hierarchy for one sample.  This starting point allows you to 

·      Begin a fully automated analysis playing with AutoGate’s fully automatic tools 
or 

·      Continue with a conventional analysis by dragging and dropping the provided gating hierarchy to other samples

2       Run a fully automated unsupervised analysis with exhaustive projection pursuit (EPP)

2.1      Select CD45+ live cells in the gating tree and then click the toolbar’s Epp button.

 

 

2.2      AutoGate responds by bringing up the following EPP configuration screen.

 

2.3      Specify the parameter for EPP to gate on

To effectively compare EPP results to the published manual results for a specific sub hierarchy we need to identify both the parameters that

1.    Were used for the manual analysis 

2.    Those that look informative for the same cell types

2.4      Point EPP at parameters used manually for CD45+ live cells

2.4.1   Click on the tree icon under the 2 parameter lists

2.4.2   Select CD45+ live cells in the gating tree

 

AutoGate populates the left list with the parameters that the OMIP investigators used in this sub hierarchy of their gating tree.

3          Point EPP at other parameters that appear informative for CD45+ live cells 

4              Click on the table icon under the 2 parameter lists

 

AutoGate shows a table of reagents with those currently used by EPP selected in a gold color.

 

5              Select parameters that have high KLD but are not already selected 

Ensure that the reagents are sorted by highest to lowest KLD (Kullback Leibler divergence).  KLD scores from 0 to 1 measure who closely a data distribution resembles the normal distribution.  0 is most resemblance and thus least interesting/informative structure.

 

 

 

 

 

Add (mac key + select / Ctrl key + select) the 6 reagents with the highest KLD that are not already selected:  Sirpa, CD40, CD95, CD32, CD85 and CD163

6          Run EPP

If your EPP configuration settings match the screenshot below then click the Ok button to run EPP!!

 

On a 6 core MacBook it takes EPP around 10 minutes to 

1.    Go through all X/Y pairings and find the top pairing where the data splits into two parts that have the best separation and are closest to each other’s size

2.    Narrow each of the two parts and repeat step 1 finding the next best data split to narrow into

3.    Stop repeating steps 1 and 2 when good 2 way splits are no longer available

EPP in progress

EPP completed

7          Compare EPP to OMIP investigator’s manual gates

When EPP completes its run it then shows the menu for running AutoGate’s matching between your fresh new EP gates and the OMIP investigator’s manual gates for the same data

7.1      Select Configure MDS & HiD match from H->2 toolbar

 

7.2      Start matching

 

Ensure your settings match as below

7.3      Evaluate matching

AutoGate shows you a resulting MDS (multi-dimensional scaling) display.  This offers many features for comparing the EPP gates to the OMIP investigator’s gates.  

 

The quickest quantitative summary can be found by clicking on the table button of the toolbar which then shows best to least match with supporting statistics.

 

 

 

7.4       Mass + Distance similarity and overlap (F- measure)

Click the highlighted buttons from the toolbar to view the graphs

 

 

7.5       Prediction Adjudicator

This helps determine how well one classification’s subsets predict another’s. PA reorganizes the predicting classifier’s subsets into predicting subsets: true positive, false positive and false negative subsets. PA determines whether the false positives or false negatives have more QFMatch based similarity to the predicted subset.

 

Click the green +- button available in the top right corner of the above table

_{PA results below}

 

7.6       Get False positive/negative rate

Click ‘See false positive/negative results’ blue button available on the top right corner of the toolbar