trajectories change a lot modifying the number of n_top_genes

[auroramaurizio]

Dear VolkerBergen, thank you for the great tool and for being always so helpful with the users! I am wondering why the trajectories in my dataset change dramatically when I modify the n_top_genes parameter in scv.pp.filter_and_normalize(adata) e.g. from 2000, to 3000 or 4000. I assume it depends on my data and I expected slightly different results playing with this parameter, but the effect is really strong. So I was wondering if you could suggest me a way to choose the best number. I am working on an integrated dataset and I was thinking to use the same number of features used for the integration (2000) and then look at the Top-likelihood genes and their shape to see if they're ok... Do you agree? Thank you! Aurora