V4.634 Separatrix Changes
1) When users select the EPP option of “best separation”, EPP looks a split between 2 parts of the data where the splitting line has the least density on the clustering grid produced by our DBM (density based merging clustering).
2) When users select the EPP option of “best balance”, EPP adjusts the split line’s density using the count of cells on each side of the split. The more equal the counts are on the 2 sides of the split the more favorable the density adjustment.
3) In v4.634 a user can select the option of “best balance+noise”. Previously EPP only counted cells on either side of the split that were within the original cluster borders before they were expanded to include the nearest unclustered cells.
Originally EPP used cell count minimums to prevent creating leaf gates that the biologist reasoned would be too small for their assay. About 4 weeks ago Wayne required that EPP use this minimum setting instead to control the creation of branch gates. As a result EPP lets the user specify the minimum be used one way or the other by selecting a checkbox “Apply to branches”
When minimum applies to leaves EPP previously only looked at the count of cells within the original cluster borders before they were expanded to include all nearest unclustered cells.
This is only correct however if the user selects the EPP option of trimming leaf gates.
For example, the settings in the panel below instruct EPP to not create a leaf gate if it has less than 50 clustered cells.
De-selecting “Trim leaf gates?” in this example instruct sEPP to not create a leaf gate if the total # of cells (clustered and unclustered) are less than
50.
But with these settings EPP would previously not create a leaf gate of 60 cells if only 49 of them were in a cluster.
But with v4.634’s improvements, EPP will now create such a 60 cell gate that has 11 unclustered cells.
In v4.634 AutoGate users in the full MATLAB environment can choose which programming code base for the separatrix they wish to use. Production users will get the new JAVA code. The JAVA is usually faster when there more than 6 clusters and becomes exponentially faster than the MATLAB as the cluster # gets higher. The C++ MEX option is likely faster for smaller cluster sizes too. I will start the JAVA to C++ conversion next week.
V4.634 improves several of the configuration setting windows by making them smaller, smoother and more consistent.
Configuration screen |
Before V4.634 |
V4.634 |
EPP configuration settings activated from the GatingTree window |
|
|
2D PathFinder configuration settings activated from PlotEditor |
|
|
See separatrix configuration settings activated from PlotEditor |
|
|
MDS Hi-D match configuration settings activated from GatingTree window |
|
|
Phenogram configuration settings activated from GatingTree window |
|
|