Datasets:

---

lucapinello

/

chorus-backgrounds

like 0

Two corrections, published together. 786 -> 753 rows.

1. CHIP transform (chorus PR #120). The 744 BPNet/CHIP rows were built with a
   per-strand softmax and expm1 of the pooled log-count. bpnet-refactor's
   orig_multi_loss=False branch — which the shipped weights uniquely pin
   (logcounts_predictions/kernel is (2,1); True gives (3,2) and refuses
   to load) — takes ONE multinomial over the flattened both-strand vector and
   builds its count target as a per-track log1p pooled with reduce_logsumexp,
   so the target is log(2 + total) and the inverse is exp(C) - 2, not
   expm1(C). Corroborated verbatim by bpnet-lite's BPNet docstring.
   Both strands now pool into one CDF row, so CHIP sample counts double to
   37,344 / 68,008 / 2,176,256. CHIP values fall (median effect ratio 1.86x);
   the old rows inflated CHIP effect percentiles by ~30% at the median.

2. Mouse removal (chorus PR #121). 33 of the 42 ATAC/DNASE rows were mm10
   models from ENCODE's mouse developmental atlas, but the builder opens
   hg38.fa and uses the hg38 DHS vocabulary — those CDFs were built by
   pushing human sequence through mouse models. Removed by pure row
   subsetting; no model was re-run.

The 9 surviving human ATAC/DNASE rows (K562, HepG2, GM12878, IMR-90 ATAC;
HepG2, IMR-90, GM12878, K562, H1 DNase) are BIT-IDENTICAL to the previous
revision, with sample counts unchanged at 18,672 / 34,004 / 1,088,128.

sha256 76f267dc862edc86052f2b25a2a8520e960dd193ad39c4ecd19e32b8a8546553
All CDFs verified finite, monotone, and non-degenerate.