Problem with Phenograph

[skylarkfi]

Hello,

I'm having problems when trying to analyze multiplex immunohistochemistry data via Phenograph. The Phenograph analysis works fine with small sample sets (e.g. 10 images), but when I try the analysis with larger sample sets (e.g. 100 images), the first ~10 resulting clusters show a cross-shape with very high values when projected onto tSNE. These clusters also get an additional empty channel (named e.g. empty_channel_102) in the channel listing. In addition, the exported .csv-files show very high marker intensities. This problem is similar to the one reported in issue #91 (https://github.com/BodenmillerGroup/histoCAT/issues/91). I've tried to run the analysis via HistoCAT GUI and also directly in Matlab, but the result is the same. I would be grateful for any advice for solving this problem. Thank you!

[skylarkfi]

Just a follow-up: when I increased the number of neighbors from 75 to 100, the Phenograph analysis succeeds. Still don't know, what causes the original problem, though.