CCR5 ZFN Benchmark (PROGNOS 2014 Extract)

⚠️ EXPERIMENTAL — the half-sites on this page are not what you pass to the CLI

The ZFN arm ships experimental in 0.6.0 with known unfixed defects tracked in #82. Do not use ZFN output for any decision without independent validation.

This page defines the (+)/(−) half-site convention the fixtures use, and that convention does not survive a round trip through sirnaforge zfn. With the default require_opposite_strands=True the CLI expects the right half-site as the reverse complement of the published (−) text, so the CCR5 pair as printed in the source tables matches no site at all, including its own on-target locus:

# published text, as tabulated below -> 0 sites
sirnaforge zfn --zfn-left-half-site GTCATCCTCATC --zfn-right-half-site AAACTGCAAAAG ...
# what actually finds chr3:46414544 -> 1 site
sirnaforge zfn --zfn-left-half-site GTCATCCTCATC --zfn-right-half-site CTTTTGCAGTTT ...

Apply the same reverse complement to every (−) half-site value in the fixtures before feeding it to the tool. See ZFN Module Guide.

This page documents the checked-in CCR5 ZFN benchmark fixtures derived from visible rows in Fine et al. (2014) supplementary extracts.

The benchmark is intentionally maintained as cleaned test data + annotation metadata, not as a loader-validation target.

Included Fixtures

  • tests/unit/data/zfn/ccr5_s10_visible_rows.csv

  • tests/unit/data/zfn/ccr5_s11_homology_visible_rows.csv

  • tests/unit/data/zfn/ccr5_benchmark_annotations.json

These fixtures preserve extraction ambiguity markers (?, Sequencing Failure) exactly as source-visible text.

S10 Coverage (Visible Rows)

ccr5_s10_visible_rows.csv includes 17 visible rows with these fields:

  • Closest gene, Match type, hg19 coordinate

  • (+) half-site, (−) half-site

  • empty/active indel and read columns

  • active mutation frequency and one-sided p-value

  • extraction notes

High-value rows captured in tests:

  • CCR5 on-target control (chr3:46414544, p = 2.7E-33)

  • CSNK1G3 novel validated off-target (chr5:123393701, active indels 17, mutation frequency 0.086%, p = 0.000019)

  • ZNF587 sequencing-failure row handling

S11 Coverage (Homology Ranking)

ccr5_s11_homology_visible_rows.csv includes 32 visible rows (rank 2 through 33) with:

  • homology rank and mismatch split (T, +, −)

  • interrogation provenance

  • closest gene, match type, coordinate, and half-sites

Tests assert:

  • rank range integrity (2…33)

  • presence of CSNK1G3 at rank 33

  • stable row-level values for deterministic regression use

Annotation Model Fixture

ccr5_benchmark_annotations.json captures reusable benchmark context:

  • nuclease/platform assumptions (CCR5-224, 4-finger monomers, heterodimeric FokI)

  • performance summary (12 prior validated off-targets, 10/12 recovery at 3×) — this is PROGNOS’s own published figure for PROGNOS, quoted from Fine et al. 2014, not a siRNAforge measurement. The JSON field is named performance.prognos_recovery_at_3x for that reason, and ccr5_study_facts.json records it as a target to regress against “once golden-site coordinates/sequences are available”. siRNAforge has never been scored against this panel, and the ZFN defects in #82 — the half-site orientation convention above, plus a default backend that cannot run the 4-mismatch budget this benchmark needs — have to be resolved before any such number would mean anything.

  • tie-break policy metadata (Exon > Promoter > Intron > Intergenic, then chromosomal location) — note the region tie-break is fed by the off-target region classifier, which #82 records as able to call a site inside a large containing gene intergenic

  • fixture scope and ambiguity markers (?, Sequencing Failure, N/A)

Notes and Limits

These fixtures are intentionally restricted to rows that are legible in extracted text. Full coordinate/sequence universes, complete rank matrices, and all validation statistics still require direct ingestion of original supplementary tables (S1, S10, S11, S14, S15).