illico is a python library performing fast and lightweight wilcoxon rank-sum tests (same as scanpy.tl.rank_genes_groups(…, method="wilcoxon")), useful for single-cell RNASeq data analyses and processing. Approximate speed benchmarks (done on a 8-CPUs, 1 GPU machine) ran on k562-essential (~300k cells, 8k genes, 2k perturbations) can be found below.

illico is compatible with python 3.11 and onward:

pip install illico -U

The full documentation is available here. A simple starter could be:

import anndata as ad
from illico import asymptotic_wilcoxon

adata = ad.read_h5ad('dataset.h5ad') # (n_cells, n_genes)
de_genes = asymptotic_wilcoxon(
       adata,
       # layer="Y", # <-- If you want tests to run not on .X, but a specific layer
       group_keys="perturbation",
       reference=["non-targeting"|None], # <- `None` computes cluster-wise DE genes. Any other `str` will be interpreted as label of the control cells.
       is_log1p=[False|True], # <-- Specify if your data underwent log1p or not
       return_as_scanpy=[False|True], # <-- Whether to return a dict compatible with Scanpy's `rank_genes_groups` function, or a pd.DataFrame holding all p-values, statistics, and fold-change
       )
# Eventually, if return_as_scanpy=True:
adata.uns["rank_genes_groups"] = de_genes

See the changelog.

The name illico is a wordplay inspired by the R package presto (now the Wilcoxon rank-sum test backend in Seurat). Aside from this naming reference, there is no affiliation or intended equivalence between the two. illico was developed independently, and although the statistical methodology may be similar, it was not designed to reproduce presto’s results.

1. scanpy also implements OVO and OVR asymptotic wilcoxon rank-sum tests.
2. pdex also implements OVO and OVR wilcoxon rank-sum tests.
3. As of March 2026, rapids-singlecell also implements OVO and OVR asymptotic wilcoxon rank-sum tests on GPU, with a focus on out-of-core datasets. If you are working with large datasets that do not fit in memory, you should check it out. For in-memory datasets, illico was benchmarked to be faster, even if CPU based.