Datasetpublic

marin-dna/zoonomia-v1-v4_utr3-order

bolinas-dna/zoonomia-v1-v4_utr3-order The bolinas-dna/zoonomia-v1-v4_utr3 cross-mammal training set, restricted to a species cohort: one representative species per NCBI order — 19 deeply-diverged placental mammals (every pair separated by ~tens of millions of years), versus the implicit-default 108 family-deduplicated species. A strict subset of the family set, so it reuses the v1 cross-mammal projection unchanged (no re-halLiftover). Same human anchors and same per-window… See the full description on the dataset page: https://huggingface.co/datasets/marin-dna/zoonomia-v1-v4_utr3-order.

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes192downloads

Dataset Card

`bolinas-dna/zoonomia-v1-v4_utr3-order`

The `bolinas-dna/zoonomia-v1-v4_utr3` cross-mammal training set, restricted to a species cohort: one representative species per NCBI order — 19 deeply-diverged placental mammals (every pair separated by ~tens of millions of years), versus the implicit-default 108 family-deduplicated species. A strict subset of the family set, so it reuses the v1 cross-mammal projection unchanged (no re-halLiftover).

Same human anchors and same per-window sequences as `bolinas-dna/zoonomia-v1-v4_utr3` — only the set of target species differs. This is a species-axis subset (a row-filter on the species column), orthogonal to the region/intervals axis that defines v4_utr3. Produced by the `snakemake/zoonomia_projection_dataset` pipeline (commit `c005cbc7a236`).

Species cohort (`order`, 19 species)

The cohort is defined by `config/species_zoonomia_447_order_dedup.tsv` — one row per species (raw HAL leaf name in the species column). Because it is a strict subset of the 108-family v1 species set, the cross-mammal projection is reused as-is; no re-halLiftover is run.

Size

2,200,430 training samples across all JSONL.zst shards — the v4_utr3 anchors projected onto the 19-species cohort, after reverse-complement augmentation. Fewer than the implicit-default 108-species `bolinas-dna/zoonomia-v1-v4_utr3` by roughly the species ratio.

Schema

Single train split of JSONL.zst shards at data/train/shard_NNNN.jsonl.zst, identical to `bolinas-dna/zoonomia-v1-v4_utr3`:

Column	Type	Description
`query_name`	str	human-window id (`win_<chrom>_<NNN>` from `windows.smk`)
`species`	str	one of the 19 cohort species
`t_chrom`	str	UCSC `chr1`-style
`t_start`	int	0-based half-open
`t_end`	int	0-based half-open; `t_end - t_start == 255`
`t_strand`	str	`+` or `-`
`t_src_size`	int	target chromosome size
`sequence`	str	exactly 255 bp; strand-aware (already RC'd if `t_strand == "-"`)
`augmentation`	str	`+` (original) or `-` (RC of `sequence`)

Construction

1.Build the v1 cross-mammal training set and its v4_utr3 region partition (see the pipeline README).
2.Filter v4_utr3 to the order species cohort via marin_dna.pipelines.projection.subset.filter_to_species (asserts the cohort is a subset of the projection's species).
3.RC-augment, shuffle (seed=42), shard to JSONL, zstd-compress, upload via hf upload-large-folder.

Source code

—Pipeline: snakemake/zoonomia_projection_dataset (latest)
—Pinned to this dataset's build: commit `c005cbc7a236`
—Species cohort list: `config/species_zoonomia_447_order_dedup.tsv`
—Base region dataset: `bolinas-dna/zoonomia-v1-v4_utr3`