Compatability with nucSeq data

[ccasar]

Hi,

this looks like a really helpful tool for my area of research! Have you tested it with nucSeq instead of scRNA data yet?

I have a liver tissue nucSeq data set where I subsetted the T cell clusters and tried to annotate them with TCAT but the result looks very diffuse:

In another liver tissue scRNA data set the results looked more like expected, i.e. more distinct clusters of annotated cells are forming.

Also another question:
The model for the multinomial label annotation seems to include only the 10 cell types shown above. I assume this is intended since in your paper you mention that for example Th1 and Th17 profiles correlated heavily with T CM/EM profiles?

Would you still suggest to use the rf_usage_normalized file to get a better idea of possible expression profile activity across the data set?

[dylkot]

Hi, thanks for reaching out. I have not tested the method out on nucSeq data. It is definitely possible there could be significant differences in the gene expression program spectra that could reduce the accuracy of representation for nucSeq. If you make plots of this using marker genes like CD4, CD8A, SELL, CCR7, FOXP3, etc. do they separate more impressively on here? If so, that would support that it is an issue with starCAT rather than the underlying data quality.

Yes, the point of the rf_usage output is to look at more diverse gene expression programs such as Th1, Th17, Th22, cell cycle and others that weren't captured in the surface markers we used to train the discrete classifier. of the 10 subsets.