ScTypePy: Python implementation of ScType

A Python implementation of ScType for automatic cell type annotation of single-cell RNA-seq data.

ScType is described in the following publication:
Nature Communications (2022): https://doi.org/10.1038/s41467-022-28803-w

Installation

pip install sctypepy

Quickstart

import scanpy as sc
from sctypepy import run_sctype

# Load and preprocess your data
adata = sc.datasets.pbmc3k()
sc.pp.neighbors(adata)
sc.tl.leiden(adata)

adata = run_sctype(adata, tissue_type="Immune system", groupby="leiden")

print(adata.obs["sctype_classification"].value_counts())

Example

Example usages can be found in the example directory.

Usage

Auto-detect tissue type

If you're unsure which tissue type to use:

from sctypepy import auto_detect_tissue_type

tissue_df = auto_detect_tissue_type(adata, groupby="leiden")
print(tissue_df.head())

Available tissue types

from sctypepy import get_available_tissues

print(get_available_tissues())

Custom marker database

Provide your own markers as a DataFrame with columns: tissueType, cellName, geneSymbolmore1, geneSymbolmore2.

import pandas as pd

custom_db = pd.DataFrame({
    "tissueType": ["Brain", "Brain"],
    "cellName": ["Neuron", "Astrocyte"],
    "geneSymbolmore1": ["SNAP25,SYT1,RBFOX3", "GFAP,AQP4,S100B"],
    "geneSymbolmore2": ["", ""],
})

adata = run_sctype(adata, tissue_type="Brain", db=custom_db)

Output

After running run_sctype():

• adata.obs["sctype_classification"] — predicted cell type per cell
• adata.obsm["sctype_scores"] — raw scores for each cell type

Authors & Contributions

• Original ScType algorithm and R implementation: Aleksandr Ianevski and contributors
  https://github.com/IanevskiAleksandr/sc-type