AIDO.Tissue Dataset Collection
niche type classification
Niche is the microenvironment in which each cell exists and is able to keep its own peculiar characteristics (Giacomo Donati, 2015). Based on spatial transcriptomic data, one can annotate niche label with established tools, which integrates similarity in gene expression profiles, spatial neighborhood structures and histological information in the tissue.
The task is to predict niche type of each cell given spatial expression data (in total 6 types). The raw dataset is from human liver sample, including a healthy sample slide. We collected the data and randomly split the total set into train, valid and split.
Each .h5ad file contains spatial coordinate information (x, y) and niche type (niche_label). The obs niche is exact name and niche_label is corresponding name index (this is input label column for running modelgenerator).
>>> import anndata as ad
>>> file = 'niche_type_classification/cosmx_liver_for_celltype_niche.test.h5ad'
>>> adata = ad.read_h5ad(file)
>>> adata
AnnData object with n_obs × n_vars = 34573 × 19264
obs: 'cellType', 'niche', 'split', 'x', 'y', 'cellType_label', 'niche_label'
>>> adata.obs
cellType niche split x y cellType_label niche_label
obs_id
c_1_103_1 Hep.5 Zone_2b test 10.25828 9.73440 1 0
c_1_103_10 Hep.6 Zone_3a test 10.61444 9.73356 7 2
c_1_103_100 Hep.1 Zone_3a test 10.57556 9.67620 3 2
c_1_103_1000 Hep.5 Zone_2a test 10.64528 9.24828 1 1
c_1_103_1001 Hep.4 Zone_2b test 10.70828 9.24180 0 0
... ... ... ... ... ... ... ...
c_1_99_995 Hep.5 Zone_2a test 8.24356 9.31284 1 1
c_1_99_996 Hep.5 Zone_2b test 8.28724 9.31428 1 0
c_1_99_997 Hep.4 Zone_2a test 8.41696 9.31512 0 1
c_1_99_998 Hep.4 Zone_2b test 8.57044 9.31524 0 0
c_1_99_999 Inflammatory.macrophages Zone_2a test 8.25976 9.31596 9 1
[34573 rows x 7 columns]
cell density
The task is to predict neighbor cell number of a target cell given the expression profiles. Neighbor cell number is counted within a specific radius. Basically, cell density distribution shows the pattern to distinguish different cell states/conditions. For instance, tumor tissue is much more aggregated than healthy tissue, thus cancer samples are more densed than normal samples.
The raw dataset is from human liver sample, including a healthy and tumor sample slide. We collected the data and curated ground truth neighbor number as in Schaar et al.. Then we randomly split the total set into train, valid and split.
Each .h5ad file contains spatial coordinate information (x, y) and density value (density).
>>> file = 'cell_density/xenium_lung_for_density.test.h5ad'
>>> adata = ad.read_h5ad(file)
>>> adata
AnnData object with n_obs × n_vars = 124058 × 19264
obs: 'density', 'split', 'x', 'y'
>>> adata.obs
density split x y
cell_id
aaaaaahk-1-0 7.0 test 497.832559 855.178702
aaaandcd-1-0 21.0 test 1732.162622 856.926639
aaabfmfn-1-0 21.0 test 1718.817560 853.211935
aaabmojc-1-0 24.0 test 1724.924030 860.927252
aaadaeog-1-0 19.0 test 2248.489685 862.145029
... ... ... ... ...
oilcikef-1-1 4.0 test 10815.658984 8476.788525
oilfbafk-1-1 5.0 test 11116.310791 8524.649170
oilfpbmk-1-1 6.0 test 11128.255615 8538.087305
oilgfkkb-1-1 2.0 test 11051.862695 8556.525830
oilgkofb-1-1 4.0 test 11085.344727 8522.173047
[124058 rows x 4 columns]
other files
scRNA_genename_and_index.tsv: gene name and index corresponds to .h5ad file
processed_fetal_lung_visium_xenium.xenium.convert.h5ad: test for dump cell embedding