Analytical validation
We check every analysis against public reference data with known truth, using objective metrics and criteria set before measuring. We publish the results — including those that do not pass — and the full methodology.
Method
For each service: a public dataset with ground truth, an objective metric, an acceptance criterion set before running, and a reproducible report (tool versions, checksums, code). If a service does not pass, it is published anyway and remains flagged as under validation.
What we have validated
Summary of the oncology, genomic, epigenomic, immunogenomic and microbiome services. Metrics against the indicated reference.
TMB · mutational burden
SEQC2 HCC1395 (tumor-normal with truth). The module matches an independent counter (±0.3%); with the production pipeline (PoN + gnomAD + VAF floor 0.05) TMB stays within 19.9% of the reference value and the negative control at 0.089.
SEQC2 · ONT/Illumina WESMSI · microsatellite instability
HCC1395 (true MSS) and HCT116 (true MSI-H): correct classification in both (0.79% vs 20.4% unstable loci).
MSIsensor-pro · SEQC2/HCT116HRD · homologous recombination
Allele-specific segments from TCGA vs BRCA1/2 mutations: separation AUC 0.979 (ovarian BRCA1/2 vs colorectal MMR) with scarHRD.
scarHRD · TCGA ABSOLUTEMutational signatures
Melanoma (UV) and colorectal (MMR): correct signature in 92.9% (UV, top-1) and 92% (MMR, by contribution) against the expected etiology.
SigProfiler · COSMIC v3.4GWAS · genomic association
Eye color in the openSNP cohort: 100% of the 5 top-hits fall within the HERC2/OCA2 locus (rs12913832, p=5×10⁻¹¹); clean null control.
PLINK2 + regenie · openSNPEpigenomics · ATAC/ChIP/WGBS
Peaks vs ENCODE GM12878: Jaccard 0.60 (ATAC) and top-1000 72% (CTCF). WGBS methylation: 82.77% vs 82.8% published by ENCODE.
MACS3 · Bismark · ENCODENeoantigens · peptide–HLA binding
Prediction against measured IC50 (IEDB, HLA-A*02:01, n=2,729): AUC 0.947 and Spearman 0.807.
MHCflurry · IEDBMicrobiome · host association
Real IBD cohort (Nielsen 2014): reproduced 4/4 IBD signature markers (depletion of F. prausnitzii q=4×10⁻²³, etc.).
MaAsLin2 · curatedMetagenomicDataLimits and context
Research (RUO)
Validations use reference data and public cohorts. Reports are delivered as research (RUO); they are not a diagnosis and do not replace accredited clinical validation.
Technical dossier
We deliver the detail per service: dataset and version, checksums, metric, criterion and result, together with the production conditions. Request it at any time.
Verifiable results, not promises
Ask for the validation dossier and we will send you the methodology and the metrics service by service.