# R4 — Public datasets and resources for building and validating SimurgArray

> **Archived Phase-0 study (sanitized copy, see `README.md`).** Written 2026-09-27, before the decisions in
> `SIMURGARRAY_MASTER_PLAN.md` §2. The analysed vendor product is called "the reference product" and its CLI
> `refcli`; its documentation URLs are replaced by `[ref-docs:…]` tokens. Where this file speaks of a clean room,
> legal clearance, or running/benchmarking the vendor binary, plan §2 supersedes it: personal research; MIT/BSD/Apache
> code may be reused with attribution; GPL tools are subprocess oracles only; vendor binaries are never run for
> comparison, inspected or decompiled; oracles are published vendor-produced GTCs, gtc2vcf and BeadArrayFiles.

Survey date: 2026-09-27. Every item marked "verified" was checked live (FTP listing, HTTP HEAD, GEO SOFT/E-utilities, or remote ZIP central-directory read), not just taken from search snippets. Items marked "reported" come from web search or documentation only.

---

## 0. TL;DR — the datasets that matter most

| Rank | Dataset | Why it matters |
|---|---|---|
| 1 | **1000 Genomes Broad Omni2.5-4v1 IDATs+GTCs, 2,141 samples** (plus BPM, 2 EGTs, CSV) | The largest fully open raw IDAT set. It includes Illumina's own GTCs to compare against GenCall, **337 complete trios**, NA12878/NA12891/NA12892, and NYGC 30x WGS truth for nearly every sample. |
| 2 | **GEO GSE165845: GSA (GSAMD-24v1/"v3"), 120 1000G samples x 3 replicates, IDAT+VCF** | Open GSA IDATs. It has triplicate reproducibility, includes NA12878, and has 30x truth. |
| 3 | **Illumina demo IDATs: GSA v4 ePGx (36 samples), GCRA ePGx (36), GDA ePGx (480), GDA v1 (36)** | These are exactly what the reference product's own tutorials run on. They are downloadable without login. The GDA v1 sheet lists **3 CEU trios (incl. NA12878) and 7 technical replicates**. |
| 4 | **GEO GSE96909 (Haraksingh 2017): NA12878 in duplicate on 9 Illumina designs** | Cross-array consistency for a single GIAB sample (OmniExpress, Omni2.5-8, Omni2.5Exome, Omni5, Omni5Exome, CoreExome, OmniExpressExome, PsychArray, CytoSNP-850K). Also a CNV gold-standard study. |
| 5 | **GIAB v4.2.1 (HG001–HG007) + NYGC 1kGP 30x (3,202 samples, 602 trios) + the vendor's public 1000 Genomes WGS reanalysis (AWS Open Data)** | Truth for SNV concordance, plus CNV/SV truth. |

---

## 1. Master table

Legend: IDAT? = raw IDAT pairs downloadable. Truth? = an orthogonal (WGS/benchmark) genotype truth exists for the samples.

| # | Name | URL | Array | n samples | Raw IDAT? | Truth? | Access / license | Use in SimurgArray |
|---|---|---|---|---|---|---|---|---|
| 1 | 1000G Omni2.5 – Broad (verified) | ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/supporting/hd_genotype_chip/broad_intensities/ | HumanOmni2.5-4v1 (B/H manifests) | 2,141 (21 pops; 337 complete trios per 1000G ped; includes NA12878, NA12891, NA12892, NA19238/39/40) | **Yes**: `Omni25_idats_gtcs_2141_samples.tgz` (106.9 GB, IDAT+GTC). Also raw and normalized intensity text (22 GB / 17 GB) | Yes: NYGC 30x, 1kGP phase 3, the reference product reanalysis. GIAB for NA12878 | Open (IGSR, Fort Lauderdale / open use) | Primary large-scale validation for normalization, GenCall re-implementation (vs bundled GTCs), clustering (GenTrain on 2k samples), LRR/BAF, Mendelian error, CNV. The directory also ships `HumanOmni2.5-4v1-Multi_B.bpm`, `…_B.egt`, `…_H.egt`, `HumanOmni2.5-4v1_B.csv` |
| 2 | 1000G Omni2.5 – Sanger (verified) | …/hd_genotype_chip/sanger_intensities/ | HumanOmni25M-8v1-1_B | 177 (79 CEU, 28 YRI, …) | **Yes**: `ALL.wgs.sanger_omni_2_5_8.20130805.snps.genotypes.idats.tar.gz` (6.2 GB). Also a GenomeStudio FinalReport (11 GB). Sanger internal IDs map via `…internal_to_coriell_id.map` | Yes (as above) | Open | A second Omni2.5 design version (8-sample chip). The GenomeStudio FinalReport gives **GenomeStudio reference output** (GType, X/Y, LRR/BAF) to diff against |
| 3 | 1000G combined Omni VCF (verified) | …/hd_genotype_chip/ALL.chip.omni_broad_sanger_combined.20140818.snps.genotypes.vcf.gz | Omni2.5 | 2,318 | No (calls only) | — | Open | Reference array calls (GRCh37) |
| 4 | GEO **GSE165845** (verified) | https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE165845 ; FTP …/geo/series/GSE165nnn/GSE165845/suppl/ | Infinium GSA (GPL23391, GSAMD-24v1-0 manifest; the series design says "GSA v3") | 120 1000G individuals (60 EUR, 60 AFR) **x 3 replicates = 360** (includes NA12878, NA19238, NA19239, HG00096, …; 91 overlap the Omni2141 set) | **Yes** (RAW.tar 45 GB; per-sample `_Grn/_Red.idat.gz` plus per-sample VCF) | Yes (NYGC 30x; GIAB for NA12878) | Open (GEO). Paper: Li et al. 2021 Genome Res, PMID 33536225 | GSA-class validation: triplicate reproducibility, concordance vs 30x, and a cross-array check vs Omni on the same 91 people |
| 5 | Illumina **GSA v4.0 ePGx Demo** (verified) | https://support.illumina.com/downloads/infinium-gsa-v4-pgx-demo-data-set-iscan.html → `…/infinium-gsa-pgx/idats_Demo_36.zip` | GSA-PGx-48v4-0_20079540_E2 | 36 (72 IDATs, 3 BeadChips; also iScan .sdf/.xml/.jpg/Metrics.txt) | **Yes** (872 MB, no login) | Sample identities not published (the "Demo_36" naming suggests a Coriell panel like the GDA sheet: unconfirmed) | Illumina website terms: RUO, "personal, non-commercial", **no redistribution** | Primary the reference product compatibility target (GSA v4 is the flagship). Run the reference product and SimurgArray on it and diff GTC/VCF/PGx outputs |
| 6 | Illumina **GCRA ePGx Demo** (verified) | `…/infinium-gcra-pgx/idats_Demo_36.zip` (linked from the reference-product downloads page) | GCRA-PGx-24v1-0 C2 | 36 (72 IDATs, 6 chips) | **Yes** | Unknown identities | Illumina terms | Second design for compatibility tests |
| 7 | Illumina **GDA ePGx Demo** (verified) | https://support.illumina.com/downloads/infinium-gda-pgx-demo-data-set-iscan.html → https://s3.amazonaws.com/webdata.illumina.com/downloads/productfiles/global-diversity-array-pgx/GDAPGx_Demo_idats.zip | GDA_PGx-8v1-0 | **480** (960 IDATs, 60 chips; per-chip `_qc.txt`) | **Yes** (16.5 GB) | Unknown identities | Illumina terms | Large-batch GDA test, and a test of clustering/normalization at scale on an 8-sample chip |
| 8 | Illumina **GDA v1.0 Demo 36** (verified) | https://webdata.illumina.com/downloads/productfiles/global-diversity-array/infinium-global-diversity-array-8-v1-0-D1-demo-data-36.zip ; sheet: https://support.illumina.com/content/dam/illumina-support/documents/downloads/productfiles/global-diversity-array/infinium-global-diversity-array-8-v1-0_D1_SampleSheet_Demo_36.csv | GDA-8v1-0_D1 | 36 = CEU Coriell. **Trios NA12878/NA12891/NA12892, NA12801/12812/12813, NA12864/12872/12873**; replicates of NA06993, NA06994, NA11920, NA11993, NA12740, NA12814 | **Yes** (465 MB) | Yes: all are 1000G/NYGC 30x samples. NA12878 has GIAB | Illumina terms | **Best small end-to-end fixture**: it has a trio (Mendelian), replicates (reproducibility) and GIAB truth, and the sample sheet already has Parent1/Parent2/Replicate columns |
| 9 | GEO **GSE96909** (Haraksingh et al. 2017) (verified) | https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE96909 (SubSeries GSE96723/789/790/791/810/816/846/893/895) | OmniExpress-24, OmniExpressExome-8, Omni2.5-8, Omni2.5Exome-8, Omni5-4, Omni5Exome-4, CoreExome-12, PsychArray-24, CytoSNP-850K (plus Affy/Agilent) | NA12878 x 2 replicates per array (30 total) | **Yes** (4.3 GB RAW.tar) | Yes (GIAB HG001; paper used a 1000G WGS CNV gold standard) | Open; BMC Genomics 18:321, PMID 28438122 | Cross-design test: the same genome across many BPMs. CNV calling benchmark; replicate concordance |
| 10 | GEO GSE178131 (verified) | GEO GSE178131 | Omni5-4 v1.1 | 4 (incl. GM12878) | Yes (839 MB) | NA12878 → GIAB | Open | Extra Omni5 NA12878 |
| 11 | GEO **GSE196829** (OneK1K) (verified) | GEO GSE196829 | GSA-24 (GPL28939) | 1,104 | Yes (11 GB RAW.tar) plus genotype matrix (12 GB) | No WGS truth | Open | Large real-world GSA cohort: clustering/QC at scale, call-rate distributions, sex checks |
| 12 | GEO GSE184150 / GSE199322 / GSE276944 (reported via esummary) | GEO | GSA | 646 / 224 / 509 | Yes (IDAT+CSV) | No | Open | Extra GSA cohorts for robustness |
| 13 | GEO GSE209804 (verified) | GEO GSE209804 | Omni5-4 v1.1 | 24 = 8 CEPH/Utah LCLs x DNA input (50 ng, 500 pg, 100 pg) | Yes (+VCF) | CEPH LCLs in 1000G (to verify per ID) | Open | Low-input / degraded-DNA stress tests (call rate, GenCall score behaviour) |
| 14 | GEO GSE19247 (verified) | GEO GSE19247 | HumanCNV370-Quad v3, HumanCytoSNP-12 v2 | 622 (single-cell MDA/WGA + bulk; **Coriell families 1463, 1423, 1990**) | Yes (2.5 GB) | Pedigree/bulk parents | Open | Pedigree (CEPH 1463 = Platinum Genomes family) on CytoSNP-12. Mendelian tests; noisy-data robustness |
| 15 | GEO GSE87838 / GSE87915 (verified) | GEO | Omni1-Quad, Omni2.5Exome-8 | 13 Illumina samples (4 trios + singleton, de novo CNV) | Yes | CNV validated in paper | Open | De novo CNV / trio CNV tests |
| 16 | Other GEO genotyping IDAT series (verified list) | E-utilities query: `IDAT[Supplementary Files] AND gse[Entry Type] AND ("SNP genotyping by SNP array"[DataSet Type] OR "genome variation profiling by SNP array"[DataSet Type])` → **254 series** | CytoSNP-12/850K, OmniExpress, Omni2.5/5, HumanHap550/610/660, 1M, CoreExome, PsychArray, MEGA, ASA, GSA | varies | Yes | Mostly no | Open | Format and design diversity (IDAT versions, old chips, 12/24/48-sample chips) for parser fuzzing and regression |
| 17 | Bioconductor **hapmap370k** (reported) | https://bioconductor.org/packages/hapmap370k | HumanHap/CNV370 Duo | 40 HapMap | Yes (in package) | 1000G/HapMap (CEU/YRI etc.) | Artistic-2.0 (Bioc) | Small, redistributable-by-license regression fixture for the IDAT reader and normalization |
| 18 | Bioconductor **IlluminaDataTestFiles** 1.50.0 (verified) | https://bioconductor.org/packages/IlluminaDataTestFiles | 370k genotyping (single Red IDAT `4019585376_B_Red.idat` + `.gz`), 450k, HT-12, WG-6 | — | Yes (tiny) | No | Artistic-2.0 | **Unit tests for the IDAT parser**: old and new formats, gzip handling. Safe to vendor in CI |
| 19 | PennCNV example (reported) | https://penncnv.openbioinformatics.org/ ; https://github.com/WGLab/PennCNV (example/) | HumanHap550 | 1 trio | No (LRR/BAF text only) | — | PennCNV license (open, non-commercial terms, check) | Expected PennCNV CNV output for the LRR/BAF → CNV stage |
| 20 | gtc2vcf docs (verified) | https://github.com/freeseek/gtc2vcf ; HapMap.md | — | — | Ships none; points to 1000G Omni FTP, GPL6986, GPL20641 | — | MIT | Reference implementation to diff against. HapMap.md gives manifests: GEO GPL20641 supplementary has `HumanOmni2.5M-8v1-1_B.bpm.gz` and `.csv.gz` (verified) |
| 21 | The vendor's public 1000 Genomes WGS reanalysis (AWS Open Data) (reported) | [ref-data:vendor-1kg-wgs-reanalysis-registry] ([ref-data:vendor-1kg-wgs-reanalysis-s3] etc.) | WGS | 3,202 (602 trios) | — | **Truth**: SNV/indel, SV, **CNV**, ROH, PGx star alleles, HLA | Open S3; license "TBD" | CNV/ROH/PGx truth matched to the reference product's own conventions. Ideal for reference-product-style CNV and PGx comparisons on the Omni2141/GSE165845 samples |
| 22 | All of Us array data (reported) | https://support.researchallofus.org/hc/en-us/articles/29475228181908 | GDA-based AoU array | ~447k | Yes (IDATs, controlled tier) | srWGS for a large subset | **Controlled** (Researcher Workbench, data use agreement; cloud-only) | Optional large-scale GDA validation if an approved researcher is involved |
| 23 | dbGaP / EGA studies (reported) | dbGaP (e.g. GENEVA/PAGE-era studies), EGA | various | various | Often yes | Sometimes | **Controlled** | Only if an application exists; not needed for core development |

---

## 2. Manifests (BPM/CSV) and cluster files (EGT)

All verified by link extraction or HTTP 200 without login unless noted.

| Array | Page | Files |
|---|---|---|
| GSA v4 ePGx (GSA-PGx-48v4-0_20079540_E2, hg38) | https://support.illumina.com/array/array_kits/infinium-global-screening-array-v4-pgx/product-files.html | `…_E2_bpm.zip` (29.5 MB, 200 OK no login), `…_E2_csv.zip`, `…_E2_ClusterFile.zip` (EGT), `…_E2_mask_file.zip`, `…_E2_pgx_model.zip`, `…_E2.zip` (bundle), `…_E2_b155_rsids.zip`, `…_E2.hg38.annotated.zip`, `GSAePGx_GS_import_v4.zip`, `DAv1.4.0-rc2-pgx-mapping-…zip`, sample sheet template, CN model and database files |
| GSA v4 (non-PGx; GSA-48v4-0_20085471_D2) | https://support.illumina.com/array/array_kits/infinium-global-screening-array/downloads.html | `GSA-48v4-0_20085471_D2.zip`, `InfiniumGlobalScreeningArrayv4.0ClusterFile-egt.zip` |
| GSA v3 (GSA-24v3-0 A1/A2) | https://support.illumina.com/downloads/infinium-global-screening-array-v3-0-product-files.html | `GSA-24v3-0-A1-manifest-file-bpm.zip`, `…-csv.zip`, `GSA-24v3-0-A1-cluster-file.zip`, A2 BPM/CSV, LIMS descriptor |
| GSA v1/v2 | https://emea.support.illumina.com/downloads/infinium-global-screening-array-v1-0-product-files.html , …-v2-0-product-files.html | BPM/CSV/EGT, annotated, rsID, coordinate-duplicate, MNV lists |
| GDA v1 (D1/D2) | https://support.illumina.com/array/array_kits/infinium-global-diversity-array/product-files.html | `https://webdata.illumina.com/downloads/productfiles/global-diversity-array/infinium-global-diversity-array-8-v1-0-D1-manifest-file-bpm.zip`, `…-csv.zip`, `…-D1-cluster-file.zip`, D2 BPM/CSV |
| GDA ePGx | https://support.illumina.com/array/array_kits/infinium-global-diversity-pgx/product-files.html | `gda+epgx-product-files-build-37.zip`, `gda+epgx-product-files-build-38-for-<product>-v1-1.zip`, the reference product model/database files, rsID files |
| GCRA ePGx | Reference-product downloads page [ref-docs:downloads] | BPM, CSV, EGT, mask, pgx_model |
| Omni2.5-4v1 | 1000G FTP broad_intensities (above) | BPM, EGT (B and H), CSV. **Hosted on the open 1000G FTP, not only on Illumina** |
| Omni2.5-8 v1.1 | GEO GPL20641 supplementary (`GPL20641_HumanOmni2.5M-8v1-1_B.bpm.gz`, `.csv.gz`); 1000G sanger dir (`HumanOmni25M-8v1-1_B.csv`, `.bpm.csv`) | BPM/CSV (no EGT) |
| HumanCNV370 | GEO GPL6986 CSV; legacy `ftp://webdata:webdata@ftp.illumina.com/…` (per gtc2vcf HapMap.md) | BPM/EGT (legacy FTP may be dead) |
| Reference-product config | Reference-product downloads page | `config_genotyping.zip`, `config_genotyping_EX.zip`, sample-sheet templates (cloud and local) |

**License and redistribution.** Illumina product files carry no explicit data license. The Illumina website legal terms (https://www.illumina.com/company/legal.html) grant viewing and downloading for "personal, non-commercial use only". They forbid modifying, reproducing, distributing or creating derivative works, and require copyright notices to be retained. Pages mark files "For Research Use Only".

What this means for SimurgArray:
- **Do not vendor or redistribute Illumina BPM/EGT/CSV or demo IDATs** in the repo, test fixtures or containers.
- Ship a **fetch script** that downloads them from the Illumina URLs above, with checksums, at test time.
- Manifests redistributed on GEO (GPL supplementary) and the 1000G FTP are publicly posted by third parties. The underlying Illumina copyright still arguably applies, so treat them the same way (download, don't bundle).
- Bioconductor test IDATs (Artistic-2.0) are the only raw-IDAT fixtures that are clearly safe to redistribute.

---

## 3. Truth sets

| Truth set | URL (verified unless noted) | Use |
|---|---|---|
| GIAB v4.2.1 small variants HG001–HG007 (GRCh37/38) | https://ftp-trace.ncbi.nlm.nih.gov/ReferenceSamples/giab/release/NA12878_HG001/NISTv4.2.1/GRCh38/ (and `AshkenazimTrio/HG002…`, `ChineseTrio/HG005…`) | Array SNV concordance inside benchmark BEDs for NA12878. Note: **GIAB has no Infinium array data** (checked HG001/HG002/HG005 data dirs). The only public HG001 IDATs are GSE96909, GSE165845, GSE178131, the GDA v1 demo and 1000G Omni. No public HG002–HG007 genotyping IDATs were found |
| GIAB HG002 SV v0.6 (Tier1) | …/AshkenazimTrio/HG002_NA24385_son/NIST_SV_v0.6/ | CNV/SV truth (HG002 only; no array IDAT for HG002) |
| Platinum Genomes 2017-1.0 (NA12877/NA12878, CEPH 1463) | https://github.com/Illumina/PlatinumGenomes ; `aws s3 cp s3://platinum-genomes/2017-1.0 . --recursive --no-sign-request` (reported) | Extra NA12878 truth with inheritance validation. The newer Platinum Pedigree (long-read, Nature Methods 2025) also covers CEPH 1463 |
| 1kGP NYGC 30x high coverage, 3,202 samples / 602 trios | https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/20220422_3202_phased_SNV_INDEL_SV/ (phased SNV/INDEL/SV per chr); pedigree `…/working/1kGP.3202_samples.pedigree_info.txt` | **Main truth for Omni2141 and GSE165845** (all samples covered). SV calls in the same VCF |
| 1kGP phase 3 SV map | https://ftp.1000genomes.ebi.ac.uk/vol1/ftp/phase3/integrated_sv_map/ | Deletion/duplication truth (GRCh37) |
| vendor 1000 Genomes WGS reanalysis CNV/SV/ROH/PGx | [ref-data:vendor-1kg-wgs-reanalysis-registry] | CNV truth in the vendor's sequencing-pipeline conventions. Used by the Genomics 2024 CNV-tool benchmark on the Omni2.5 IDAT set |
| 1000G pedigree (phase 3) | ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/working/20130606_sample_info/20130606_g1k.ped | Trio definitions for Mendelian error (337 complete trios within Omni2141, verified) |

---

## 4. Strand / annotation / reference resources

| Resource | URL | Notes |
|---|---|---|
| Will Rayner strand files | https://www.chg.ox.ac.uk/~wrayner/strand/ (old: well.ox.ac.uk, which redirects here); ilmnStrand/, sourceStrand/ subdirs; RefAlt files at http://mccarthy.well.ox.ac.uk/2016/07/files-to-update-refalt-on-illumina-chips/ | Per-chip, per-build (b36/b37/b38) `.strand`, `.miss`, `.multiple` and RefStrand/RefAlt files. Covers GSA versions, Omni, OmniExpress, CytoSNP, GDA (reported). The site returned CloudFront 403 to automated access, so check it in a browser. No explicit license: use as an **independent cross-check** of SimurgArray's own probe-sequence-to-reference alignment (TOP/BOT → REF/ALT), not as a shipped dependency |
| Illumina annotated/rsID files | per-array product pages (b150/b153/b155 rsIDs, hg38 annotated) | Locus-to-rsID mapping as the reference product uses it |
| GRCh37 (1000G) | ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/reference/human_g1k_v37.fasta.gz | gtc2vcf default |
| GRCh38 no-alt analysis set | ftp://ftp.ncbi.nlm.nih.gov/genomes/all/GCA/000/001/405/GCA_000001405.15_GRCh38/seqs_for_alignment_pipelines.ucsc_ids/GCA_000001405.15_GRCh38_no_alt_analysis_set.fna.gz | Public domain (NCBI) |
| dbSNP (b155/b156) | https://ftp.ncbi.nlm.nih.gov/snp/latest_release/VCF/ | Public domain; rsID annotation |
| Illumina BeadArrayFiles | https://github.com/Illumina/BeadArrayFiles | BSD-2. GTC/BPM/EGT format reference (no test data) |
| Illumina GTCtoVCF | https://github.com/Illumina/GTCtoVCF | Apache-2.0, semi-archived. Reference for VCF conventions (no test data) |
| freeseek gtc2vcf | https://github.com/freeseek/gtc2vcf | MIT. IDAT→GTC (GenCall-like), GTC→VCF, BAF/LRR |

---

## 5. Trio and replicate datasets with IDATs (Mendelian error and reproducibility)

| Dataset | Trios | Replicates | Array |
|---|---|---|---|
| 1000G Omni2141 | **337 complete trios** (ped-based, verified) | Some cross-center duplicates (Broad vs Sanger overlap) | Omni2.5-4 / -8 |
| GSE165845 | none complete (only partial, e.g. NA19238/NA19239) | **120 x 3 technical replicates** | GSA |
| Illumina GDA v1 Demo_36 | **3 CEU trios** (incl. NA12878) | 6 replicate pairs | GDA-8v1-0 |
| GSE96909 | — | NA12878 x2 on each of 9 designs | multiple |
| GSE19247 | Coriell families 1463/1423/1990 (bulk plus single-cell) | many single-cell replicates | CNV370 / CytoSNP-12 |
| GSE87838/87915 | 4 clinical trios | — | Omni1 / Omni2.5Exome |
| GSE209804 | CEPH | 8 samples x 3 input amounts | Omni5-4 |

---

## 6. Published benchmarks: the reference product vs GenomeStudio vs gtc2vcf

- **No independent, peer-reviewed head-to-head benchmark of the reference product vs GenomeStudio vs gtc2vcf genotype calls was found.**
- Illumina's reference-product documentation (release notes, output-file docs) says the outputs are "concordant" with previous solutions but gives no numbers. It also documents convention differences: the reference product orients BAF/LRR to reference-genome REF=A/ALT=B, while GenomeStudio is reference-agnostic, so BAF can appear as 1 − GS value. See [ref-docs:output-files] . SimurgArray must replicate this orientation.
- The gtc2vcf README makes no quantitative GenomeStudio comparison.
- The closest public evidence is below.

| Study | URL | Key facts |
|---|---|---|
| Establishing analytical validity of BeadChip array genotype data by comparison to WGS and benchmark datasets (2022) | https://pmc.ncbi.nlm.nih.gov/articles/PMC8919546/ | GSA-24 v1.0, 263 Coriell samples in triplicate (789 assays), GenomeStudio 2.0.2 plus AutoConvert/GenCall, `gtc_to_vcf.py` 1.1.1. Concordance 0.9932 (GSA) vs 0.9981 (WGS) against 1KG. WGS BAMs in SRA PRJNA792997. **IDATs not deposited.** Its metrics are a template for SimurgArray's validation protocol |
| A systematic benchmark of CNV detection tools for high-density SNP arrays (Genomics 2024) | https://www.sciencedirect.com/science/article/pii/S0888754324001836 | Uses the **1000G Omni2.5 IDAT set (2,141 samples)** processed in GenomeStudio 2.0, with the vendor 1000 Genomes WGS reanalysis as CNV truth. PennCNV was best on precision/F1. Directly reusable design for SimurgArray's CNV module |
| Haraksingh et al. 2017 BMC Genomics | https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-017-3658-x | NA12878 on 17 array designs (IDATs in GSE96909). CNV benchmark vs 1000G WGS gold standard |
| COPILOT (containerised Illumina genotyping QC) | https://www.biorxiv.org/content/10.1101/2021.07.26.453753 | GenomeStudio-based QC protocol; useful QC metric definitions |

**Implication.** SimurgArray's own validation will be the first public three-way comparison. Recommended protocol:
1. Run *(superseded: vendor-binary runs are not part of the plan, see plan §2; use published vendor-produced GTCs instead)* the reference product (local, free with Illumina login), GenomeStudio/IAAP-CLI, gtc2vcf and SimurgArray on the Illumina demo IDATs (GSA v4, GDA) and on the open GSE165845 and Omni2141 sets.
2. Report per-locus GT concordance, GenCall score correlation, LRR/BAF RMSE, call rate, and concordance to NYGC 30x / GIAB in benchmark regions.
3. Report Mendelian error rate (Omni trios, GDA demo trios) and replicate concordance (GSE165845 triplicates, GDA demo replicates).

---

## 7. Practical download plan (sizes)

1. **CI / small (under 1 GB):** Bioconductor IlluminaDataTestFiles (25 MB). GDA v1 Demo_36 (465 MB) plus its BPM/EGT. GSA v4 ePGx Demo (872 MB) plus the E2 BPM/EGT (fetched, not vendored).
2. **Medium:** GSE96909 (4.3 GB), GCRA demo, Sanger Omni 177 (6.2 GB; includes a GenomeStudio FinalReport for exact-output diffs).
3. **Large:** GSE165845 (45 GB), GDA ePGx demo (16.5 GB), Broad Omni2141 IDAT+GTC (107 GB; its GTCs give Illumina AutoCall reference output for 2,141 samples).
4. **Truth:** GIAB v4.2.1 HG001 (GRCh37/38); NYGC 3202 phased panel (only the relevant chromosomes/samples via `bcftools view -s`); 1000G ped; vendor 1000 Genomes WGS reanalysis CNV VCFs for the overlapping samples.

## 8. Gaps

- No public HG002–HG007 Infinium IDATs were found (GEO search for NA24385/NA24143/NA24149/NA24631 returned only methylation arrays).
- Demo-set sample identities for the GSA v4 / GCRA / GDA-ePGx demos are not published. Infer them by genotype matching against the NYGC 30x panel (a cheap first SimurgArray task).
- The Rayner strand site blocks scripted access (CloudFront 403); fetch it manually.