implementing a obs.col based splitting

[BioinfoTongLI]

Hi @Claptar , just wondering how difficult it is to implement this kind of splitting here?
https://github.com/cellgeni/scraft/blob/develop/src/scraft/cli/split_h5ad.py

[Claptar]

Hi @BioinfoTongLI ! I'm a bit reluctant to implement some complex behaviour like conditional filtering, but I will surely consider it if I get more requests for this feature. For now I suggest to use csvkit for any pandas/sql-like filtering. Here is a small example.

1. Export cell metadata to a CSV file:

uv run h5ad export dataframe visium.h5ad obs --output cells.csv
wc -l cells.csv # 2689 cells.csv

2. Get all cluster names from cells.csv:

awk -F ',' 'NR>1{print $4}' cells.csv | sort | uniq -c

284 Cortex_1
257 Cortex_2
244 Cortex_3
164 Cortex_4
129 Cortex_5
226 Fiber_tract
222 Hippocampus
208 Hypothalamus_1
133 Hypothalamus_2
105 Lateral_ventricle
42 Pyramidal_layer
68 Pyramidal_layer_dentate_gyrus
153 Striatum
261 Thalamus_1
192 Thalamus_2

3. Get all obs names in "Cortex_2" with csvsql from csvkit:

csvsql -d ',' -I --query "SELECT _index FROM cells WHERE cluster='Cortex_2'" cells.csv > barcodes.txt
sed -i '1d' barcodes.txt # remove header
wc -l barcodes.txt  # 257 barcodes.txt

4. Use this list to create a subset .h5ad file:

uv run h5ad subset visium.h5ad --output cortex2.h5ad --obs barcodes.txt

5. Check the result:

uv run h5ad info cortex2.h5ad

An object with n_obs × n_var: 257 × 18078
        obs:    array_col, array_row, cluster, in_tissue, leiden, log1p_n_genes_by_counts, log1p_total_counts, log1p_total_counts_mt, n_counts, n_genes_by_counts, 
pct_counts_in_top_100_genes, pct_counts_in_top_200_genes, pct_counts_in_top_500_genes, pct_counts_in_top_50_genes, pct_counts_mt, total_counts, total_counts_mt
        var:    feature_types, gene_ids, genome, highly_variable, highly_variable_rank, log1p_mean_counts, log1p_total_counts, mean_counts, means, mt, n_cells, n_cells_by_counts, 
pct_dropout_by_counts, total_counts, variances, variances_norm
        obsm:   X_pca, X_umap, spatial
        varm:   PCs
        obsp:   connectivities, distances
        uns:    cluster_colors, hvg, leiden, leiden_colors, neighbors, pca, rank_genes_groups, spatial, umap