Single Cell ssGSEA is an extension of the Single Sample Gene Set Enrichment Analysis (ssGSEA) method[1] for use with single cell RNA-sequencing (scRNA-seq) data. Because of the sparsity of scRNA-seq data, ssGSEA scores computed in individual cells are subject to uncertainty. The Single Cell ssGSEA approach reduces this uncertainty by:
For more details and benchmarking results, see this preprint.
Single Cell ssGSEA is available as a Docker image and as a PyPI Python package. This method is also available as a GenePattern module on the GenePattern Cloud Server.
To install the Docker image, run docker pull genepattern/sc_ssgsea.
To install the Python package, run python -m pip install sc-ssGSEA.
Note: Single Cell ssGSEA accepts multiple input formats (see below). To run Single Cell ssGSEA on a Seurat object saved in an .rds file, R and Seurat must be installed in the same environment. The Single Cell ssGSEA Docker image has both installed, but users of the PyPI package will need to install R and Seurat separately.
Create an account on the GenePattern Cloud Server and search for the “Single Cell ssGSEA” module.
Parameters
input.file : A file containing a Seurat object in RDS format, an AnnData object in H5AD format, or a Seurat object in H5Seurat format. The file must end in .rds, .h5ad, or .h5seurat respectively.
gene.sets : The gene sets to test for enrichment. Choose one or more options from either the Human or Mouse collections listed in the dropdown.
cluster.data.label : The name of the column in the metadata contained in input.file which contains a grouping of cells.
chip.file : Optionally, provide a .chip file to transform genes into a different namespace or convert between human and mouse orthologous genes.
output.file : The name of the file that contains the results matrix. By default, this is scores. Input will have the suffix .tsv appended.
Please note that H5AD files must contain untransformed counts. Please see Using H5AD files for more information.
The following code calls Single Cell ssGSEA on an RDS file containing a Seurat object, assuming that seurat_object.rds is the file, which contains a metadata column called seurat_clusters, and that gene sets are defined in gene_sets.gmt.
from sc_ssGSEA import read_gmt, run_ssgsea_parallel, Expression
## Load and parse expression + cell labels
expr = Expression.get_expression_object(
"seurat_object.rds",
"seurat_clusters"
)
expr.load()
## Load gene sets
gs, _ = read_gmt("gene_sets.gmt")
## Run single cell ssGSEA
sc_ssGSEA_scores = run_ssgsea_parallel(
expr.metacells,
gs,
)
The first argument to Expression.get_expression_object() can be an RDS file containing a Seurat object, H5AD file containing an AnnData object, or an H5Seurat file containing a Seurat object. Single Cell ssGSEA relies on the file suffix, which must be either .rds, .h5ad, or .h5seurat respectively.
Please note that H5AD files must contain untransformed counts. Please see Using H5AD files for more information.
If you wish to use another file format, create a Python class that inherits from Expression and implements the load() method. The load() method may assume that it has access to the fields _filepath: str and _group_name: str (the metadata column), and it must populate the fields _gene_names: List[str], _cell_names: List[str], and _group_labels: List[str]. load() should then create a scipy.sparse.csr_matrix containing the expression data, call Expression._normalize_sparse_matrix(sparse_mat: csr_matrix) and then populate _metacells: pandas.DataFrame using the functionExpression._get_metacells(sparse_mat: csr_matrix).
While Scanpy workflows often involve transformations to the raw counts populating the AnnData object, Single Cell ssGSEA requires untransformed integer counts and will expect them in adata.raw field. Users should call adata.raw = adata.copy() prior to any normalization or other numerical transformations. Single Cell ssGSEA will check that adata.raw contains integer counts by sampling some expression values and checking that they are equivalent to the nearest integer value within a small tolerance threshold.