no highly variable genes

I ran the newest Scanpy package’s

sc.pp.highly_variable_genes(adata, min_mean=0.0125, max_mean=3, min_disp=0.05,
                            batch_key='batch')

It indeed gave me information about highly_variable_nbatches etc. But all the genes were labelled as not variable (‘False’).

Any ideas?

[LuckyMD]

I just checked again… and it’s not exactly the same… if you select n_top_genes then, you will get the top genes shared by the most batches. If you select thresholds for mean and dispersion, you will use these thresholds against the mean dispersion and mean mean across all batches. And those can be lower than the thresholds if HVGs are not shared between many batches. So to be safe, you can go with selecting n_top_genes.

[brianpenghe]

OK. So If I understand it correct, when batch_key is used, adata.var[‘highly_variable’] is just adata.var[‘highly_variable_genes_intersection’] ?