CoolFace
Datasetpublic

marin-dna/zoonomia-v1-v4_ccre_noexon_enhancer

bolinas-dna/zoonomia-v1-v4_ccre_noexon_enhancer A curated enhancer training set for issue #326 — a de-contaminated derivation of the v4 ccre_non_promoter arm of bolinas-dna/zoonomia-v1-v1, built by the snakemake/zoonomia_projection_dataset pipeline at commit 6b320c268547. Provenance This subset is v4_ccre_noexon further restricted to enhancer-dominant windows (dELS+pELS basepair coverage ≥ the other non-PLS cCRE classes), population-matching the val_enhancer… See the full description on the dataset page: https://huggingface.co/datasets/marin-dna/zoonomia-v1-v4_ccre_noexon_enhancer.

sourceHugging Faceupdated 3mo agoView on Hugging Face
0likes1.1kdownloads
Dataset Card

bolinas-dna/zoonomia-v1-v4_ccre_noexon_enhancer

A curated enhancer training set for issue #326 — a de-contaminated derivation of the v4 ccre_non_promoter arm of `bolinas-dna/zoonomia-v1-v1`, built by the `snakemake/zoonomia_projection_dataset` pipeline at commit `6b320c268547`.

Provenance

This subset is v4_ccre_noexon further restricted to enhancer-dominant windows (dELS+pELS basepair coverage ≥ the other non-PLS cCRE classes), population-matching the val_enhancer LL-gap probe and removing the ~17% non-enhancer (CA / CTCF / TF) heterogeneity (#283 H2). Filter: region_labels.select_ccre_noexon_enhancer.

It is a strict subset of `bolinas-dna/zoonomia-v1-v4_ccre_non_promoter`; concatenating the curated subsets does not reconstruct v1 (they are experimental probes, not a partition). Contains 62,734,288 training samples (human anchors × up to 108 Zoonomia mammals × reverse-complement augmentation) — the exact row count across all JSONL.zst shards.

Schema

Identical to `bolinas-dna/zoonomia-v1-v1` — a single train split of JSONL.zst shards at data/train/shard_NNNN.jsonl.zst; key field sequence is exactly 255 bp, strand-aware. See the parent card for the full column table.

Source code