GMM for identification of positive cells

[LazDaria]

Hi,

thank your for your work!
In your paper, you describe the identification of positive cells based on GMMs. Afterwards, do you binarize your data (e.g. setting signal to 1 and background to 0) or do you use some kind of normalization of positive and negative cells? How does GMM work if there are no positive cells for a given marker and sample, does it only find one mixture in that case?

Thanks in advance!

[afrendeiro]

Hi there!

Thanks for the interest.

we ran the analysis in two steps: first we did model selection with GMMs of various mixtures (2-6) and selected the best based on a metric (please check methods).
On a second step, we binarized the data (0/1 classification) with the best model by classifying the top two mixtures as positive (or top if a model of 2 mixture was selected).

Importantly, this was run for all cells of all images at once (~350,000) which helped make sure every marker had positive cells.

Overall this is likely an inferior approach to using marker intensity as continuous values, so use with care.

Best wishes,
André