input expression matrix

[wangjiawen2013]

Dear, In the case of diffxpy (eg wald test, de.test.wald), what the input matrix should be ? raw counts or normalized data ? And, I found that diffxpy ran very slowly when the dataset is large ( 10,000 cells and 20,000 genes ), then how to speed up it ?

[davidsebfischer]

Hi @wangjiawen2013, this depends on the noise model that you want to use. For most scenarios, I recommend negative binomial noise and for that you should provided unprocessed counts, ie integer values that come out of the aligner. Let me know if that is unclear. We are working on the speed issue, we are introducing new estimation backends this week, I will comment again in this issue then! Would you mind sharing your diffxpy call, or just your formula_loc? Thanks! Best, David