subcellular-proteomics-curation

An adjusted version of CZI's cellxgene-schema, repurposed for subcellular proteomics datasets. The original tool validates single-cell genomics data against the cellxgene schema using GENCODE gene references; this fork swaps gene validation for UniProt protein validation and trims the genomics-only machinery (ATAC-seq, spatial/Visium, species mapping, schema migration, and the original CLI).

It is used as a library enabling validation scripts for anndata objects that are generated by grassp analysis.

Installation

The package lives in the cellxgene_schema_cli/ subdirectory. Install it directly from GitHub with pip:

pip install "git+https://github.com/czbiohub-sf/subcellular-proteomics-curation.git#subdirectory=cellxgene_schema_cli"

This pulls in all dependencies, including the ontology data package grassp-ontology-guide (a CZ Biohub fork of cellxgene-ontology-guide that adds non-animal organisms such as yeast). That wheel is referenced directly in requirements.txt; if you need to install it on its own:

pip install https://github.com/czbiohub-sf/grassp-ontology-guide/releases/download/v1.9.0/cellxgene_ontology_guide-1.9.0-py3-none-any.whl

Install from a local clone

make install   # runs: cd cellxgene_schema_cli && pip install -r requirements.txt && pip install .

Usage

The package is consumed as a library. The key entry points:

from cellxgene_schema import validate
from cellxgene_schema.write_labels import AnnDataLabelAppender
from cellxgene_schema.uniprot import GeneChecker

# Validate an in-memory AnnData object
validator = validate.Validator()
validator.adata = adata
validator._set_schema_def()
validator._deep_check()
print(validator.errors, validator.warnings)

# Append human-readable labels (protein names, lengths, locations, ontology labels)
appender = AnnDataLabelAppender(adata)
appender._add_labels()

See validate_objects.py and annotate_objects.py in the curation_grassp project for the full validation and annotation workflow.

Protein references

Protein validation uses gzipped UniProt TSV files bundled in cellxgene_schema_cli/cellxgene_schema/uniprot_files/. Supported organisms are configured in the SupportedOrganisms enum in cellxgene_schema/uniprot.py. The reference files are regenerated with cellxgene_schema_cli/scripts/protein_processing.py --force, and outdated UniProt accessions can be remapped with scripts/update_uniprot_ids.py.

Bundled reference version

The files currently committed were downloaded from UniProt release 2026_02 (released 10-June-2026), retrieved 26-August-2026.

file	accessions	reviewed (Swiss-Prot)
proteins_homo_sapiens.tsv.gz	210,709	20,431
proteins_mus_musculus.tsv.gz	87,782	17,267
proteins_rattus_norvegicus.tsv.gz	61,540	8,232
proteins_drosophila_melanogaster.tsv.gz	42,895	3,899
proteins_saccharomyces_cerevisiae.tsv.gz	6,733	6,733

Each file has the columns requested by the per-species query in uniprot_files/protein_info.yml: Entry, Reviewed, Length, Gene Names (primary), Gene Names (synonym), Annotation (annotation score), Protein names, Subcellular location [CC].

Schema definition

The validation rules live in cellxgene_schema_cli/cellxgene_schema/schema_definitions/schema_definition.yaml.

Development

make unit-test   # run the unit tests
make check       # pre-commit (black, ruff) + mypy

Authors

This is a fork of CZI's single-cell-curation (the cellxgene-schema tool), originally developed by the Chan Zuckerberg Initiative.

This fork is maintained by the Computational Biology Platform at the Chan Zuckerberg Biohub San Francisco.