mantispy.pp.rank_int#
- mantispy.pp.rank_int(adata, by=None, c=0.375, stochastic=True, seed=0, key_added=None, copy=False)[source]#
Replace every feature by the normal quantile of its rank.
- Parameters:
- adata
AnnData Object to transform. Run it after
normalize(), as the JUMP recipe and the batch-correction benchmark do.- by
str|None(default:None) Rank within each group of this column.
Noneranks globally, as the reference implementation does, which keeps every feature comparable across the screen. Ranking per plate also removes plate-level differences in distribution shape, but can hide a plate that failed.- c
float(default:0.375) Blom’s constant.
- stochastic
bool(default:True) Tie handling; see
rank_inverse_normal.- seed
int(default:0) Tie handling; see
rank_inverse_normal.- key_added
str|None(default:None) Write to
layers[key_added]instead of overwritingX.- copy
bool(default:False) Return a modified copy instead of transforming in place.
- adata
- Return type:
- Returns:
None, or the modified copy. WritesXorlayers[key_added].
Notes
Every feature comes out standard normal, so no feature dominates a distance through its units. Effect sizes are lost as well: a feature that doubled and one that moved by one percent look the same if they reorder the same wells. Keep the untransformed values with
key_addedfor effect sizes and dose-response curves.