Jackstraw method computed on genes rather than cells

[nilseling]

Hi,

could it be that the 'expr_values' matrix needs to be transposed here:
https://github.com/RubD/Giotto/blob/fefab049ce9747acd8a15d75ed67daa1ab16ebd3/R/dimension_reduction.R#L187

I guess the jackstraw permutations are performed on the genes rather than the cells at this point.
Best,

Nils

[RubD]

Hi Nils,

Thanks for your comment.

Here we wanted to identify which PCs (based on genes) should be used to cluster cells downstream. In order to identify stable (or significant) PCs the jackstraw method randomly and iteratively permutes a set of genes and then computes how this affects each PC. So following this logic it seems that permutations on genes is correct, right?

Best,
Ruben

[nilseling]

Hi Ruben,
hmm, maybe I'm confused. For me the jackstraw method only returns a comparable result to signPCA(gobject, method = "screeplot") when transposing the matrix. But I only had a quick look and might have to think about this further.
Best,
Nils