Closed Andrew-InTheBox closed 6 years ago
FYI - some additional info, these are exported files from FCS Express exported to keep a subset of cells of interest from gating live and dump channel negative. It's an 18 color experiment so lots of spreading error and biexponential transformation is necessary. It seems the transformations weren't applied to the exported values and somehow this was doing something unexpected when clustering, for example some clusters show mean values of negative infinity.
I used R integration in FCS Express to save csv files of the same data, with biexponential transformations applied, and the clustering in vortex now proceeds as expected. Perhaps values below 0 were set to 0 or at some point during data import or clustering divided by 0, thus the negative infinity values, and only for some rare events which may be why clustering on a random subset of 10k works.
I'm attempting to cluster the data in the files in the drive location below and regardless of how I import the data or set the clustering configuration (I've tried importing them from fcs or csv versions of the data), I get an odd result where there are always 8 clusters, with all events in the 1st cluster except for 1 event per cluster in the remaining 7 clusters, so it seems very artifactual. However, if I do a random subset of 10k events, it seems to work fine.
For import, I just import the area parameters for all fluorescent and scatter parameters (Except the FSC-405 parameter). I use transform value = 10 and scale to 1SD.
https://drive.google.com/drive/folders/1QccS10kzLlpw13sEEVRn4gYRz9mwVedK?usp=sharing
Edit - seems more of an issue with the data than with Vortex, see below.