using CITE-seq data to cluster single cells

Question

[chi.delta]

Hello,

I am trying to cluster single cells using the ADT reads from the CITEseq (generated with the 10x platform). Ideally, I would like to use the information from two ADT features. I have tried the kmeans function on a count matrix of the two antibodies in order to make 3 clusters (number pre-defined) but this doesn't seem to work. Is there any way to do that? Any suggestion would be appreciated. Thanks a lot!

[ATpoint]

Think about it, how would any clustering with only two genes/features be informative? Is this really what you need? If you break it down, for two features the combinations can basically be low-low, low-high, high-low and high-high, given that these combinations really exist in the data. Also, what does "not work" mean?

[chi.delta]

Let's say that based on known FACS data, I already know that three populations exist based on the two antibodies staining: single positive for each and double positive. However, I do not want to arbitrarily just set a threshold of what is considered positive and negative in the cite-seq reads of the two antibodies, and also because after normalization, the bimodality of the staining is lost, I am looking for a reliable way to define the populations above in my cite seq data.

[ATpoint]

If it is really just two genes then why not just plotting logcounts of gene1 vs logcounts of gene2 as in a FACS scatter plot and see whether there is any sort of separation. While automated methods may or may not work you could simply define a few gates by hand or at least decide if there is any separation at all.