Datasets:

netbiolab
/

MetacellFormer-corpus

Tags:

Dataset Preview

Split (1)

train

The full dataset viewer is not available (click to read why). Only showing a preview of the rows.

The dataset generation failed

Error code:   DatasetGenerationError
Exception:    ArrowInvalid
Message:      Failed to parse string: 'TCATTACTCTGGAGAG' as a scalar of type int64
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1848, in _prepare_split_single
                  writer.write_table(table)
                  ~~~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 784, in write_table
                  self._write_table(pa_table, writer_batch_size=writer_batch_size)
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 795, in _write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2312, in cast_table_to_schema
                  cast_array_to_feature(
                  ~~~~~~~~~~~~~~~~~~~~~^
                      table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                      feature,
                      ^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
                  return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
                                           ~~~~^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2152, in cast_array_to_feature
                  return array_cast(
                      array,
                  ...<2 lines>...
                      allow_decimal_to_str=allow_decimal_to_str,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
                  return func(array, *args, **kwargs)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2015, in array_cast
                  return array.cast(pa_type)
                         ~~~~~~~~~~^^^^^^^^^
                File "pyarrow/array.pxi", line 1186, in pyarrow.lib.Array.cast
                File "/usr/local/lib/python3.14/site-packages/pyarrow/compute.py", line 414, in cast
                  return call_function("cast", [arr], options, memory_pool)
                File "pyarrow/_compute.pyx", line 604, in pyarrow._compute.call_function
                File "pyarrow/_compute.pyx", line 399, in pyarrow._compute.Function.call
                  result = GetResultValue(
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                  return check_status(status)
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
                  raise convert_status(status)
              pyarrow.lib.ArrowInvalid: Failed to parse string: 'TCATTACTCTGGAGAG' as a scalar of type int64
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1879, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

source_cell_id int64	metacell_id int64
0	240
1	192
2	241
3	76
4	219
5	21
6	82
7	100
8	29
9	29
10	49
11	76
12	96
13	16
14	16
15	242
16	210
17	199
18	183
19	43
20	281
21	291
22	96
23	181
24	73
25	225
26	202
27	71
28	25
29	291
30	82
31	265
32	209
33	249
34	280
35	265
36	276
37	71
38	126
39	93
40	234
41	170
42	298
43	82
44	300
45	43
46	82
47	164
48	233
49	103
50	21
51	193
52	119
53	173
54	307
55	62
56	273
57	193
58	254
59	135
60	74
61	81
62	38
63	278
64	202
65	190
66	38
67	29
68	124
69	261
70	138
71	62
72	82
73	308
74	93
75	29
76	111
77	122
78	193
79	103
80	308
81	62
82	291
83	82
84	240
85	111
86	82
87	34
88	252
89	257
90	93
91	120
92	100
93	101
94	300
95	21
96	288
97	126
98	49
99	234

End of preview.

MetacellFormer pretraining corpus

This repository provides the 818,609 metacell profiles used for MetacellFormer pretraining, constructed by summing raw counts from 44,389,000 cells across 1,334 source datasets. It also provides the source-cell memberships and fixed training–validation split: 799,562 metacells for training and 19,047 for validation. The matrices preserve the gene features and expression values used in the study.

File	Contents
`source_manifest.tsv`	Source accessions, citations, releases, terms, sizes and relative file paths
`metacells/{dataset_id}.h5ad`	Sparse raw-count sums in `X`; gene IDs and symbols in `var`; constituent cell counts and split in `obs`
`memberships/{dataset_id}.tsv.gz`	`source_cell_id`, `metacell_id`, in the selected source-cell order
`splits/{dataset_id}.tsv.gz`	`profile_id`, `split`, in metacell row order; `train` or `eval` for the tokenizer
`checksums.sha256`	SHA-256 file checksums

IDs are unique within a source dataset; use (dataset_id, ID) across sources. eval in the split files means validation. Metacell matrices are unnormalized sums, not mean expression or log-transformed values. Keep all supplied features when computing the normalization denominator.

Use

After installing the GitHub package in MetacellFormer-env, choose a row of source_manifest.tsv and use its metacell_file, split_file and dataset_idx:

python scripts/tokenize_training.py \
  --input CORPUS/metacells/DATASET_ID.h5ad \
  --split CORPUS/splits/DATASET_ID.tsv.gz \
  --dataset-id DATASET_ID --dataset-idx DATASET_INDEX \
  --model-dir MODEL_DIRECTORY --output-dir tokens/DATASET_ID

MODEL_DIRECTORY supplies the vocabulary, reference medians and fixed value-bin edges. Check downloaded files with sha256sum -c checksums.sha256.

Reconstructing source counts

Full single-cell matrices are obtained from the original providers. The releases are CELLxGENE Census 2025-11-08 and scBaseCount 2026-01-12. Source selection and MetaQ membership are already fixed; do not draw a new sample or rerun MetaQ to reproduce this corpus.

CELLxGENE: source_cell_id is the decimal row index of the dataset-specific filtered Census export, not a Discover barcode or a Census soma_joinid. Use the Census release above, X_name="raw", the exact source_filter in the manifest, and retain the complete Census gene axis with feature_id. The original export used cellxgene_census.get_anndata with observation metadata excluding soma_joinid, producing sequential string row IDs. Preserve this row order and check its row count against source_input_cells. A current Discover download is not interchangeable with this view.
scBaseCount: source_cell_id is the source H5AD observation name. The manifest links the pinned Gene/Homo_sapiens file. Follow the provider's access instructions if its bucket requires an authenticated billing project.

Prepare the selected source matrix and aggregate using the published memberships:

python scripts/prepare_corpus_source.py --input upstream_counts.h5ad \
  --corpus-root CORPUS --dataset-id DATASET_ID --output selected_counts.h5ad
python scripts/aggregate_metacells.py selected_counts.h5ad \
  CORPUS/memberships/DATASET_ID.tsv.gz reconstructed.h5ad --source-id DATASET_ID

The preparation command checks cell IDs and the frozen gene universe. Specify --counts-layer if needed. It reads chunks; the selected sparse output remains in memory. The aggregation command sorts metacell IDs lexically, so compare reconstructed and released matrices after alignment by metacell ID and gene ID.

Attribution and terms

CELLxGENE-derived data retain CC BY 4.0; scBaseCount data retain CC0 1.0. Original source citations and terms are recorded per dataset in the manifest. Changes comprise the recorded cell selection, metacell raw-count summation, fixed split and reduced metadata. See LICENSES.md.

Downloads last month: 27

Total file size:

62.9 GB

source_cell_id int64	metacell_id int64
0	240
1	192
2	241
3	76
4	219
5	21
6	82
7	100
8	29
9	29
10	49
11	76
12	96
13	16
14	16
15	242
16	210
17	199
18	183
19	43
20	281
21	291
22	96
23	181
24	73
25	225
26	202
27	71
28	25
29	291
30	82
31	265
32	209
33	249
34	280
35	265
36	276
37	71
38	126
39	93
40	234
41	170
42	298
43	82
44	300
45	43
46	82
47	164
48	233
49	103
50	21
51	193
52	119
53	173
54	307
55	62
56	273
57	193
58	254
59	135
60	74
61	81
62	38
63	278
64	202
65	190
66	38
67	29
68	124
69	261
70	138
71	62
72	82
73	308
74	93
75	29
76	111
77	122
78	193
79	103
80	308
81	62
82	291
83	82
84	240
85	111
86	82
87	34
88	252
89	257
90	93
91	120
92	100
93	101
94	300
95	21
96	288
97	126
98	49
99	234

source_cell_id int64	metacell_id int64
0	240
1	192
2	241
3	76
4	219
5	21
6	82
7	100
8	29
9	29
10	49
11	76
12	96
13	16
14	16
15	242
16	210
17	199
18	183
19	43
20	281
21	291
22	96
23	181
24	73
25	225
26	202
27	71
28	25
29	291
30	82
31	265
32	209
33	249
34	280
35	265
36	276
37	71
38	126
39	93
40	234
41	170
42	298
43	82
44	300
45	43
46	82
47	164
48	233
49	103
50	21
51	193
52	119
53	173
54	307
55	62
56	273
57	193
58	254
59	135
60	74
61	81
62	38
63	278
64	202
65	190
66	38
67	29
68	124
69	261
70	138
71	62
72	82
73	308
74	93
75	29
76	111
77	122
78	193
79	103
80	308
81	62
82	291
83	82
84	240
85	111
86	82
87	34
88	252
89	257
90	93
91	120
92	100
93	101
94	300
95	21
96	288
97	126
98	49
99	234

MetacellFormer pretraining corpus

Contents

Use

Reconstructing source counts

Attribution and terms

source_cell_id int64	metacell_id int64
0	240
1	192
2	241
3	76
4	219
5	21
6	82
7	100
8	29
9	29
10	49
11	76
12	96
13	16
14	16
15	242
16	210
17	199
18	183
19	43
20	281
21	291
22	96
23	181
24	73
25	225
26	202
27	71
28	25
29	291
30	82
31	265
32	209
33	249
34	280
35	265
36	276
37	71
38	126
39	93
40	234
41	170
42	298
43	82
44	300
45	43
46	82
47	164
48	233
49	103
50	21
51	193
52	119
53	173
54	307
55	62
56	273
57	193
58	254
59	135
60	74
61	81
62	38
63	278
64	202
65	190
66	38
67	29
68	124
69	261
70	138
71	62
72	82
73	308
74	93
75	29
76	111
77	122
78	193
79	103
80	308
81	62
82	291
83	82
84	240
85	111
86	82
87	34
88	252
89	257
90	93
91	120
92	100
93	101
94	300
95	21
96	288
97	126
98	49
99	234