cellink.tl.external.run_seismic_torch

cellink.tl.external.run_seismic_torch(adata, magma_file, cell_type_col, species='human', layer=None, min_genes=250, min_cells=50, magma_gene_col='GENE', magma_z_col='ZSTAT', device='cpu', prefix=None, save_results=True)

Run the sparse-torch Seismic backend end to end.

Drop-in alternative to run_seismic() (the R/seismicGWAS subprocess backend) that never densifies the genes x cells matrix, so it scales to atlases with millions of cells. Returns a DataFrame with the same cell_type / pvalue / FDR columns as run_seismic(), so downstream code (e.g. plotting, Cauchy combination) is backend-agnostic.

Parameters:

• adata (AnnData) – Single-cell data. species="mouse" is not auto-translated (unlike the R backend) — pass an adata already indexed by the target species’ gene symbols if you need cross-species gene mapping.

• magma_file (str or Path) – MAGMA .genes.out file with columns magma_gene_col/magma_z_col.

• cell_type_col (str) – Column in adata.obs with cell-type labels.

• layer (str, optional) – adata.layers key to use as expression. Defaults to adata.X.

• device (str, default "cpu") – torch device. Sparse ops here are CPU-friendly; only use “cuda” for very large atlases where a GPU is free.

• prefix (str, optional) – If given and save_results, associations are written to {prefix}_associations.tsv.

Return type:

DataFrame

Returns:

pd.DataFrame Columns: cell_type, pvalue, beta, se, FDR.