Validation and quality

Trust is earned by method, before results.

We share our validation approach, our testing discipline and our limits openly. The value of a genomic report lies as much in what it says it does not know as in what it knows.

Comparison with reference samples.

We test our pipeline against international reference samples whose genotypes are independently established.

  • GeT-RM reference samples

    Our pharmacogenomic calls are independently compared with reference genotypes from the CDC’s Genetic Testing Reference Material program.

  • Public reference genomes

    End-to-end runs are performed on public samples with known genotypes, such as Genome in a Bottle HG002.

  • Pre-registered blind checks

    Clinically critical regions such as DPYD and SLCO1B1 are fixed in advance and checked before results are seen.

  • Stage-by-stage error tracing

    A missed indel or call is traced until we find the stage where it was lost; the fix becomes a permanent test.

Validation results

Concordance rates and sample counts will be published on this page, with figures, once the independent assessment is complete.

Testing discipline.

Every change to our software passes automated tests before it touches clinical data.

  • More than 2,600 automated tests

    The analysis engine and interface are tested with thousands of automated tests on every change; past bugs are kept as regression tests.

  • A fingerprint for every run

    Every analysis is recorded with its job ID, timestamps and SHA-256 digests of its input files. The result proves which data it came from.

  • No silent assumptions

    Data with an unverifiable genome build is not converted; without a kit selection, analysis never falls back to a default; an interrupted run never resumes on its own.

  • One source for every verdict

    Guideline verdicts come from a single version-controlled knowledge base; even our custom modules’ calls are passed through it.

  • No call is not “reference”

    For every locus without a result, the measured depth and the reason are written in the report.

  • Document integrity

    Every report is sealed with a SHA-256 digest that proves its content has not been altered.

Our limits.

Every method has limits. We state ours in the report and here.

  • Short-read data

    CYP2D6 copy-number changes and hybrid alleles are assessed with limited resolution; they are measured by depth ratio and a confirmatory test (MLPA or long-read) is recommended when needed.

  • Structural variants

    Some structural variants and copy-number states cannot be resolved reliably with short reads; such cases are stated in the report.

  • Non-star-allele variants

    The functional effect of variants found by loss-of-function screening is often unverified; these findings never change the phenotype and are presented separately.

  • Clinical context

    Concomitant drugs, organ function and clinical status can change a genotype-based verdict. The final decision always rests with the physician.

Regulatory status.

The BIODECODE pharmacogenomic report is a clinical decision-support document that helps physicians with drug selection and dosing. It does not replace diagnosis, clinical history or the physician’s judgement. It is for research use and is not IVD-approved.

Sample intake, laboratory analysis and medical sign-off are performed by the responsible medical genetics diagnostic centre, İstanbul Genetik Grubu. Bioinformatic analysis, interpretation and report generation are BIODECODE’s responsibility.