cosine shift
I am trying to replicate the figures in the paper and follow the examples for the in silico perturbation.
When trying to model cell states, the code calculates cosine similarity shift (i.e. delta) between the cell states before and after the permutation, and returns this delta value as cosine shift.
However for gene perturbation it also returns a cosine_shift metric (cosine_shift_mean/stdev etc). But as far as I can tell this is not the shift/delta amount (higher means more perturbation), but the actual cosine similarity (lower means more perturbation). Is that correct, is it just a column naming issue, or is there a delta calculation for genes as well that I missed?
Thank you for your comment! Yes, these are cosine similarities - we updated the column names accordingly to clearly indicate this.