FASTQ in — aligned BAM, called VCF, QC report, and a signed Ed25519 provenance manifest out. Compute + artifacts only. The lab stays the CLIA/CAP lab of record.
What every per-report deliverable contains
One signed, audited bundle per sample. Verifiable on the lab's hardware in five seconds.
genomics-verify CLI.Reproducible numbers, published methodology
HG002 30× WGS DeepVariant via Parabricks 4.7.0-1 against GIAB v5.0q, scored under the same hap.py methodology as the PrecisionFDA Truth Challenge V2.
HG002 30× against GIAB v5.0q, scored with hap.py + GIAB v3.6 stratifications. Aggregate F1 hides where a caller is strong vs weak. This is the table your QMS reviewer asks for.
| Region | SNP F1 | Indel F1 | Notes |
|---|---|---|---|
| In-scope (complement of exclusion BED) | 0.9993 | 0.9959 | Headline number; covers MHC, PAR, all easy + medium-difficulty |
| MHC / HLA region (in-stratum) | 0.9897 | 0.9498 | Lifted in-scope per ADR-0006; hardest region we report on |
Difficult regions (alldifficultregions union) |
0.9521 | 0.9308 | Segdups + tandem repeats + low-mappability + homopolymers; out-of-scope per QUALITY_CLAIMS |
| Aggregate v5.0q (raw, no exclusion) | 0.9906 | 0.9408 | Whole-genome aggregate; published for transparency |
| Aggregate v4.2.1 (legacy truth set) | 0.9954 (combined) | Industry-standard short-read baseline | |
Full per-cell hap.py CSVs, exclusion-BED methodology, and the ADR-0006 MHC-lift rationale are in the Quality Metrics dossier.
Two independent runs of the locked Parabricks digest on the same HG002 30× FASTQ produced byte-identical variant calls: 7,593,649 of 7,593,649 variants, zero delta. Variant-content SHA-256 matched bit-for-bit across both runs.
Raw BAM/VCF bytes differ only in Parabricks' embedded
@PG header (output filename + run timestamp);
no alignment-record or variant-call content varies between runs. Verify the claim yourself
by running scripts/run_repeatability_proof.sh
on your own GB10. Full proof JSON in
the diligence package.
In-scope numbers are scoped to the complement of the published exclusion BED
(uncompressed SHA-256 7dc4d16b…3932c1c;
GIAB v3.6 alldifficultregions minus MHC
∪ chrX/Y non-PAR/XTR/ampliconic). The MHC / HLA region is in scope.
Numbers are within ±0.0005 of the top short-read submissions in the
PrecisionFDA Truth Challenge V2 — Illumina DRAGEN v4.x (anonymized
submission W607K), Sentieon DNAscope (KXBR8), Google DeepVariant Open
Source (BSODP) — re-scored under identical hap.py / vcfeval methodology
and identical GIAB v3.6 stratifications (Olson 2022,
DOI 10.5281/zenodo.6384789). Definitive numbers and methodology archive
are in the customer bundle.
How it works
Step 1
Globus, GCS BYOB, signed URL, or SFTP. ~70 GB pair for a typical 30× WGS sample.
Step 2
Parabricks DeepVariant on GPU. ~2 hours on a single GB10; faster on burst capacity. Container digest pinned.
Step 3
BAM + VCF + QC report + JCS-canonical signed manifest + audit-log entries. Every SHA-256 bound by signature.
Step 4
One command. No network. Cross-check against the fingerprint pinned in the executed pilot agreement.
$ genomics-verify manifest.json signature.json --public-key genomics-public.pem
OK — signature valid for this manifest.
algorithm: ed25519
public key id: <fingerprint pinned in the pilot agreement>
manifest sha256: <hex>
job_id: <uuid>
sample_id: <your sample id>
outputs: 3 file(s)Pay per sample. No box capex.
Free pilot for the first 5–10 samples while the Lab validates concordance against its existing pipeline.
$0/ sample
5–10 samples, free. The Lab keeps the side-by-side concordance report regardless of outcome.
$149/ WGS sample
All-inclusive per-report price. WES from $79.
$249/ WGS sample
Same-day turnaround SLA with a dedicated reserved GPU.
Enterprise: volume discounts, custom SLAs, dedicated GCS project, white-label deployments. Available for labs at 200+ samples / month. Talk to sales →
We are compute + artifact delivery. The Lab is the lab of record.
We deliver secondary analysis only: FASTQ → BAM, VCF, gVCF, QC report, signed provenance manifest, audit-log entries.
The customer laboratory is the CLIA/CAP lab of record and performs all clinical interpretation, variant classification, report sign-out, and patient communication. We do not diagnose, do not classify pathogenicity, do not sign out reports, and do not recommend treatment.
Quality numbers cited above are conditional on the platform staying inside its locked intended use and the published exclusion BED. Customer written acknowledgement of the exclusion BED is required before any GIAB v5.0q-scoped quality claim is published for that Lab.
5 to 10 samples, $0. We return signed BAM/VCF/QC plus a side-by-side concordance report against your existing pipeline. The dossier walks through every artifact. No commitment.
Or email us directly at hello@datamade.ai.