Adding completely two different cell population at the end time point in GLM (Generalized Linear Model)?

Question

[mmchowdhury9]

Hi guys,

Hope you all are doing well.

I have counts data from samples of cell reprogramming in two conditions at different time points. At last time point, I collected two different cell population samples for each condition, but for other time points only one cell population sample. When I include all sample data in GLM, #DEGs is very much lower than #DEGs I obtain when I include only one cell population sample at the last time point. Note that, one population at the last time point is quite similar to the samples collected at previous time points. I am not sure whether is it okay to use two population at one time point in GLM in edgeR. And also as one population is very similar to the previous population, whether inclusion of it is causing some dampening effect thereby reducing #DEGs. If this is true, then, is it alright only to use one population at the end time point in GLM?

I would be greatful if you could give some insights in this. Thanks in advance.

Cheers,

Mahfuz

[mmchowdhury9]

a spelling mistake in the question title corrected.

Answer 1

[Kevin Blighe]

You should check via PCA bi-plots how the cell-types relate to each other after having applied different models.

I would first check (PCA bi-plot) without adjusting for cell-type, and then see how well the cells overlap, mainly on PC1, and also how they overlap in relation to time-point. If you notice any biasing between cell-type in this way, then you could adjust for them via the GLM.

[mmchowdhury9]

Thanks so much Kevin.