zoonomia-v1-v2

Same projection backbone as bolinas-dna/zoonomia-v1-v1, subsetted to windows whose human anchor overlaps [TSS − 256, TSS + 256] for any Ensembl rel 115 protein_coding transcript. ~7.1% of v1.

Produced by the zoonomia_projection_dataset pipeline in Open-Athena/bolinas-dna — permalinked at the exact code that built this dataset: snakemake/zoonomia_projection_dataset @ 7ff07cd (PR #158).

Row count

Train: 15,863,610 rows.

Verified by zstd -dc data/train/*.jsonl.zst | wc -l over all 64 shards (58 shards × 247,869 + 6 shards × 247,868, np.array_split-balanced). About 7.1% of zoonomia-v1-v1's 223,880,280 rows.

The HuggingFace dataset viewer happens to show the same 15,863,610 here (full parquet conversion, partial: false), but it's unreliable in general — for the sibling zoonomia-v1-v1 it reports off-by-22% estimates and a partial-conversion 14M, neither matching the true 224M (see #166). The number above is hand-verified.

How intervals were constructed

Same projection backbone as bolinas-dna/zoonomia-v1-v1:

1. Tile hg38 (Ensembl rel 115) into 255 bp windows (bedtools makewindows -w 255 -s 128, autosomes + X + Y, drop N-overlapping).
2. Conservation filter: proportion_conserved >= 0.20 against phyloP_447m (threshold calibrated against phyloP_241m >= 2.27; NaN counted as 0).
3. Cross-mammal projection: halLiftover --noDupes onto 108 family-deduped mammals; per-species collapse to single span; drop pre-resize lengths outside [128, 512]; midpoint-resize to 255 bp.
4. Strand-aware sequence extraction (bedtools getfasta -s).
5. Reverse-complement augmentation (augmentation ∈ {+, -}).
6. Shuffle (seed=42).

v2-specific filter — protein-coding mRNA TSS proximity. Extract TSSes for every Ensembl rel 115 transcript with transcript_biotype "protein_coding", expand each to a [TSS − 256, TSS + 256] band, merge, and bedtools intersect -u the human anchor BED against the merged band. Keep only projected rows whose query_name belongs to a window touching at least one protein-coding TSS region. Implemented in bolinas.projection.tss.write_mrna_tss_band_bed.

Schema

Identical to bolinas-dna/zoonomia-v1-v1:

column	type	notes
query_name	str	Human window id, win_<chrom>_<NNN>
species	str	One of 108 mammals
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 reverse-complemented if t_strand == "-")
augmentation	str	+ (original) or - (RC of sequence)

Single train split. JSONL.zst shards at data/train/shard_NNNN.jsonl.zst (64 shards).

2-axis versioning

Dataset names follow zoonomia-{pipeline_version}-{intervals_version}:

• pipeline_version (here v1) snapshots species set, conservation cutoff, alignment backend, and resize/length-filter params.
• intervals_version (here v2) names this post-projection subset: protein-coding mRNA-TSS proximity. v1 is identity — see bolinas-dna/zoonomia-v1-v1.

Loading

from datasets import load_dataset

ds = load_dataset("bolinas-dna/zoonomia-v1-v2", split="train", streaming=True)

Source

• Repo: Open-Athena/bolinas-dna
• Pipeline (commit-pinned): snakemake/zoonomia_projection_dataset @ 7ff07cd
• Pipeline README: README.md @ 7ff07cd
• Producing PR: #158