Third-party attribution

Chorus wraps eight deep-learning oracles and one genome-browser library. Each ships under its own license, and model weights are not redistributed in this repo — they are fetched from the original authors' hosts at first-use time.

Deep-learning oracles

Oracle	Authors	Paper	Weights / code license
Enformer	Avsec et al., DeepMind	Effective gene expression prediction from sequence by integrating long-range interactions (Nature Methods 2021)	Apache-2.0 (code); weights on TensorFlow Hub
Borzoi	Linder et al., Calico Labs	Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation (Nature Genetics 2025)	Apache-2.0 (code); weights on Zenodo
ChromBPNet	Pampari et al., Kundaje Lab (Stanford)	ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility (bioRxiv 2024)	MIT (code); weights on ENCODE
Cherimoya / CATv1	Schreiber	Preprint forthcoming — cite the repository: github.com/jmschrei/cherimoya	MIT (code); CATv1 weights CC-BY-4.0, fetched lazily per experiment from programmable-genomics/CATv1
Sei	Chen et al., Troyanskaya Lab (Princeton)	A sequence-based global map of regulatory activity for deciphering human genetics (Nature Genetics 2022)	BSD-3-Clause (code + weights)
LegNet	Penzar et al., Vaishnav Lab (Broad)	LegNet: a best-in-class deep learning model for short DNA regulatory regions (Bioinformatics 2023)	MIT (code); weights bundled with source
AlphaGenome	Avsec et al., Google DeepMind	AlphaGenome: advancing regulatory variant effect prediction with a unified DNA sequence model (Nature 2026)	Gated on HuggingFace (google/alphagenome-all-folds); accept the license to download weights

Chorus does not modify the upstream model code beyond the adapter layer in chorus/oracles/<name>.py. Each oracle's predict / score semantics are those of the original publication.

EPInformer-seq is deliberately absent from that table: it is not third-party. Its weights are trained for chorus and served from this project's own HuggingFace repo (lucapinello/chorus-epinformerseq-v2), there is no vendored upstream code under chorus/oracles/epinformerseq_source/, and the architecture is described in chorus/oracles/epinformerseq.py. What it does borrow is the ChromBPNet-style frozen bias net that subtracts Tn5/MNase sequence preference in logit space — credited to the Kundaje lab in the row above. Recorded here because an audit reasonably read "7 of 8 oracles attributed" as a missing attribution rather than as a first-party model (2026-08-12 audit, F5). If the name is meant to credit upstream EPInformer work, add that citation here — this note deliberately does not invent one.

Bundled third-party JavaScript

• IGV.js (Integrative Genomics Viewer, Robinson et al., Broad/UCSD) — igv.org, github.com/igvteam/igv.js, MIT license. Shipped as chorus/analysis/static/igv.min.js (1.3 MB, inlined) so a report needs no CDN for the library itself. Source license at github.com/igvteam/igv.js/blob/master/LICENSE.

  A report no longer resolves its genome through igv.org's hosted registry (#139): chromosome lengths, the ideogram and the gene track are bundled, which took one report from 14 requests across two hosts to 9 across one, and — on one report, the SORT1 Cherimoya panel — from 9.6 s to 2.2 s to paint (across the whole 19-report corpus the range moved 8.6–10.8 s to 2.2–4.4 s, a 3.9x mean). One resource is still remote — the reference sequence (hg38.2bit from UCSC), because every igv.js version requires a sequence source and hg38 is 3 GB. Point CHORUS_IGV_SEQUENCE_URL at a self-hosted copy and a report needs no internet at all; serving it same-origin with the report measured 0 external requests and 0.8 s.

• Bundled inside igv.min.js — the igv.js UMD bundle vendors third-party libraries of its own, which chorus therefore ships too. Their licence banners survive inside the file (grep -o '@license[^*]*' chorus/analysis/static/igv.min.js):

  library	notice in the bundle	licence
  DOMPurify	@license DOMPurify 3.2.1 | (c) Cure53 and other contributors	Apache-2.0 OR MPL-2.0
  pako	@license (MIT AND Zlib)	MIT AND Zlib

  Noted because the audit asked whether the bundle "carries its upstream license header": the UMD bundle opens with code rather than a banner, so the licences are present but not at the top of the file, and these two were credited nowhere outside it (2026-08-12 audit, F6).

• UCSC hg38 cytoband table (cytoBandIdeo, primary chromosomes only) — shipped as chorus/analysis/static/cytoBandIdeo_hg38.txt.gz, 6.1 kB, from hgdownload.soe.ucsc.edu/goldenPath/hg38/database/cytoBandIdeo.txt.gz. UCSC genome-annotation data is free to use and redistribute (genome.ucsc.edu/license). igv.js draws the ideogram from it, and the per-chromosome maximum band end supplies the chromosome lengths, so this one file replaces both the cytoBandIdeo and the hg38.chrom.sizes fetch.

• GENCODE v48 basic annotation supplies the inline gene track, scoped to the drawn window (~59 genes and 5.5 kB for a 1 Mb locus). It replaces the UCSC ncbiRefSeq track the registry used to attach, and has the advantage of being the same annotation chorus uses for every gene lookup, so the panel agrees with the numbers printed beside it. Already credited under the annotation section below; downloaded, not vendored.

Per-track background CDFs

The NPZ CDFs under <data-dir>/backgrounds/ are derived from the oracle authors' published predictions on a reference set of genomic loci. They are computed by Chorus and distributed at huggingface.co/datasets/lucapinello/chorus-backgrounds under CC-BY-4.0 — attribute the original oracle publications above when citing numbers derived from them.

Chorus itself

MIT-licensed (see LICENSE). Cite as:

Pinello Lab. Chorus: unified interface for genomic deep-learning oracles. 2026.