Title: Intended Use Statement — Genomics Compute Layer Version: 0.2.0-draft Status: Draft (§1.3 v5.0q exclusion BED published) Owner: Head of Quality / Regulatory Last Reviewed: 2026-05-06 Next Review: 2026-08-06 # Intended Use Statement This document is the **locked** intended-use statement for the {{COMPANY_NAME}} genomics compute layer (the "Platform"). The intended-use statement is the load-bearing boundary that keeps the Platform out of the FDA Software-as-a-Medical-Device (SaMD) regulatory path and out of the direct CLIA/CAP scope. Every other document in this dossier, every customer-facing communication, and every product feature must respect this boundary. If a proposed feature, claim, or contract term would violate this statement, it must be escalated through the change-control SOP **before** implementation, marketing, or customer commitment. ## 1. Scope ### 1.1 In Scope The Platform performs **secondary bioinformatics analysis** on de-identified or pseudonymous sequencing data submitted by the customer laboratory. Specifically, the Platform: - Accepts paired-end FASTQ inputs (and supporting reference + index files) from the customer's storage location. - Executes a pinned Parabricks container (default `nvcr.io/nvidia/clara/clara-parabricks:4.7.0-1`, exact digest pinned in `technical/PIPELINE_LOCK.md`) on a GPU worker (local GB10, or Brev- provisioned cloud GPU when queue depth justifies burst). - Produces aligned BAM, VCF, gVCF, run logs, a per-run QC summary, and a cryptographically-checksummed manifest containing the provenance fields enumerated in §1.4. - Returns job state and artifact locations to the submitting lab via the Platform API. ### 1.2 Out of Scope (Explicit) The Platform does **not** perform, and the {{COMPANY_NAME}} service offering does **not** include, any of the following: - **Clinical interpretation** of variants (pathogenic / likely pathogenic / VUS / likely benign / benign assignment). - **ACMG/AMP variant classification** or any equivalent clinical-grade classification framework. - **Clinical report generation or sign-out.** No PDF or human-readable clinical report is produced. - **Diagnosis or treatment recommendation** of any kind. - **Primary clinical decision-making** for a patient, including triage, risk stratification, or therapy selection. - **Tertiary analysis** that draws clinical conclusions (e.g. tumor / normal somatic interpretation, HRD scoring with clinical claims, PRS reporting with clinical claims). - **Direct-to-patient or direct-to-consumer** delivery of any artifact. - **Acting as the laboratory of record** for any test. The Platform is **not a medical device**, **not a SaMD**, **not a clinical laboratory**, and **not an LDT**. {{COMPANY_NAME}} does **not** hold a CLIA certificate and does **not** seek CAP accreditation under this product scope. ### 1.3 Reportable-Range Exclusions The Platform makes **no quality claim** for the following regions or variant classes. Customers must explicitly exclude these from their internal reportable range, or independently validate them, before using the Platform output for any reportable result. | Exclusion | Reason | Tracking | |-----------|--------|----------| | GIAB v5.0q HG002 30x exclusion BED (`investigations/v5_0q_excluded_regions.bed`, SHA-256 (uncompressed) `7dc4d16b1d0eb1d171713bc272c9a3f3b881dddb1f305faba02dac25a3932c1c`, 4,571,604 intervals, 747,356,696 bp; post-MHC-lift per ADR-0006). Specifically: GIAB v3.6 `alldifficultregions` **minus MHC** ∪ chrX_nonPAR ∪ chrX_XTR ∪ chrX_ampliconic ∪ chrY_nonPAR ∪ chrY_XTR ∪ chrY_ampliconic. PAR regions remain in scope; the MHC / HLA region (chr6:28,510,020-33,480,577) is in scope per ADR-0006. | Per-stratum decomposition (`V5_0Q_GAP_ANALYSIS.md` §5.10) shows that 97.3 % of the 121,994 v5.0q FNs (118,748) lie inside this BED after MHC lift. The in-scope complement (notinalldifficultregions ∪ MHC ∩ chr1-22 ∪ PAR) operates at SNP F1 = 0.9993 and Indel F1 = 0.9959 on HG002 30x (arithmetic estimate per ADR-0006; hap.py confirmation pending), exceeding the v4.2.1 aggregate F1 of 0.9954. v5.0q's assembly-based extension goes almost entirely into difficult-region territory; this is a denominator change by NIST, not a platform regression. | `investigations/V5_0Q_GAP_ANALYSIS.md` §5.10, §6; `decisions/0006-mhc-exclusion-lift.md`; `investigations/v5_0q_excluded_regions.bed` | | HaplotypeCaller-derived outputs | The current HaplotypeCaller benchmark output is empirically invalid (Precision = NaN, Sensitivity = 0.0000 in both v4.2.1 and v5.0q runs surveyed). HaplotypeCaller is excised from the locked pipeline; HaplotypeCaller results must not be used as a quality basis until the benchmark is fixed and HaplotypeCaller is reintroduced through change control. | `investigations/HAPLOTYPECALLER_BENCHMARK_FIX.md`; `technical/PIPELINE_LOCK.md` §10b | | Difficult-to-map regions (segmental duplications, centromeres, telomeres, decoy / alt contigs) | Standard short-read limitation; not a Platform-specific claim. | GIAB v3.6 stratifications (`genome-stratifications-GRCh38@all.tar.gz`, MD5 `f6b0666987b01aec08e9f5fa17707b10`, SHA-256 `c5a1eceac54aac2c438af21825223d2a71e64b3db6b1c9e923849babb38063d8`). Specific BEDs: `Union/GRCh38_alldifficultregions.bed.gz`, `SegmentalDuplications/GRCh38_segdups.bed.gz`, `OtherDifficult/GRCh38_MHC.bed.gz`, `Mappability/GRCh38_lowmappabilityall.bed.gz`. Operationalized as the exclusion BED at `investigations/v5_0q_excluded_regions.bed` for the v5.0q evaluation. | | Structural variants > 50 bp | DeepVariant short-variant caller is not validated for SV. SV is out of scope until a separate validation is performed. | `validation/` (separate bundle) | | Copy-number variation | CNV is out of scope of this product; not produced by the pipeline. | n/a | | Mitochondrial variants | `[TBD] ` | n/a | | Sex chromosomes: chrX_nonPAR, chrX_XTR, chrX_ampliconic, chrY_nonPAR, chrY_XTR, chrY_ampliconic. PAR is **in scope** (autosomal-equivalent, diploid genotypes). | Sex-chromosome non-PAR/XTR/ampliconic content is excluded (when sex is declared) per GIAB v3.6 XY stratifications; caller behavior on sex chromosomes outside PAR depends on ploidy hint and is not validated as a clinical claim. | `investigations/v5_0q_excluded_regions.bed` (incorporates the chrX/Y XY strata listed); GIAB v3.6 `XY/GRCh38_chr{X,Y}_{nonPAR,XTR,ampliconic}.bed.gz`. | | Any non-GRCh38 reference build | Validation evidence is on GRCh38 only; T2T-CHM13 and GRCh37 are out of scope. | n/a | | Inputs below the supported QC floor (read length, coverage, quality) | Listed in `technical/PIPELINE_LOCK.md`; out-of-spec inputs return WARN/FAIL QC and no quality claim is made. | `[TBD] ` | ### 1.4 Required Per-Run QC and Provenance Every successful job produces, at minimum: - **QC metrics**: FASTQ checksum, BAM checksum, read count, FASTQ quality summary, alignment rate, mean and median coverage, percent of bases above configured coverage thresholds, duplicate rate, insert-size summary, contamination estimate, sex / ploidy consistency check, callable-region summary, variant counts by type, Ti/Tv ratio, Het/Hom ratio, runtime, per-test compute cost, and an overall PASS / WARN / FAIL with reason codes. - **Provenance fields**: input file paths and SHA256 checksums, exact command line executed, container image digest, host metadata (hostname, kernel, CUDA driver, GPU model and count), GPU compute capability, queue / start / end timestamps, output file checksums, and absolute log path. Reason codes, thresholds, and the PASS/WARN/FAIL decision logic are defined in `technical/QC_SPEC.md`. ## 2. Intended Users The Platform is intended to be used **only** by trained personnel acting on behalf of a CLIA-certified (and, where applicable, CAP-accredited) clinical laboratory. Specifically: - **Bioinformatics staff** of {{CUSTOMER_LAB}} who configure pipeline parameters, submit jobs, and review per-run QC. - **Clinical operations staff** of {{CUSTOMER_LAB}} who route artifacts into the laboratory's downstream interpretation and reporting workflow. - **Laboratory directors** of {{CUSTOMER_LAB}} (CLIA-defined high- complexity laboratory director) who hold ultimate responsibility for the clinical use of any artifact produced. The Platform is **not** intended for use by: - Patients or research subjects directly. - Treating clinicians directly (clinicians receive interpretation from the laboratory of record, not from the Platform). - Researchers using the Platform output for publication-grade clinical claims without their own independent validation. ## 3. Intended Environment The Platform is intended to be operated in support of a customer-validated CLIA / CAP laboratory environment. Specifically: - Customer laboratory must hold a current CLIA certificate appropriate to the test complexity and, where applicable, CAP accreditation. - Customer laboratory must perform its own assay validation (or verification, for an FDA-cleared test) before reporting any clinical result derived from Platform output. - Customer laboratory must operate the Platform under its own quality management system (QMS), with documented SOPs covering at minimum: job submission, QC review, artifact custody, incident handling, and reportable-range definition. - Customer laboratory must execute a Business Associate Agreement (BAA) with {{COMPANY_NAME}} before submitting any data subject to HIPAA. The BAA template is in `customer/` (separate bundle). The Platform itself runs on: - Local **NVIDIA GB10** GPU host operated by the customer or by {{COMPANY_NAME}} on behalf of the customer under contract. - **Brev**-provisioned cloud GPUs (e.g. B300) when queue depth justifies a burst, subject to the burst policy in `technical/ARCHITECTURE.md`. GPU compute capability requirements and the supported host configuration list are pinned in `technical/PIPELINE_LOCK.md`. ## 4. Required Disclaimers Every customer-facing artifact, manifest, API response, and customer-facing document must carry the following disclaimer (or the version then-current in `intended-use/INTENDED_USE.md` and `intended-use/QUALITY_CLAIMS.md` — a dedicated `customer/DISCLAIMERS.md` is not maintained; the language below is the authoritative customer-facing disclaimer): > The {{COMPANY_NAME}} genomics compute layer is a secondary-analysis > compute and artifact-delivery service. It is **not** a medical device, > **not** a clinical laboratory, and **not** a clinical interpretation > service. Outputs are not clinical results. The customer laboratory is > solely responsible for assay validation, reportable-range definition, > clinical interpretation, and report sign-out under its own CLIA > certificate and (where applicable) CAP accreditation. The disclaimer must be visible in: - The manifest JSON (`disclaimer` field). - The Platform API `/jobs/{id}` response (`disclaimer` field). - Every customer-facing PDF, web page, marketing email, and sales deck. - The customer onboarding packet. ## 5. Phased Commercial Path Alignment This intended-use statement applies to all four phases of the commercial path. It does **not** change between Phase 1 (paid validation pilot), Phase 2 (parallel production comparison), Phase 3 (lab internal sign-off), and Phase 4 (paid per-test clinical production compute). The Platform's boundary is the same in all phases. What changes between phases is the customer's confidence in the Platform's output, not {{COMPANY_NAME}}'s regulatory posture. In particular: **Phase 4 production compute does not make {{COMPANY_NAME}} a clinical laboratory.** The customer is the lab of record in Phase 4 exactly as in Phase 1. ## 6. Change Control Any change to this document — including any addition to §1.1 (In Scope), any removal from §1.2 (Out of Scope), any narrowing of §1.3 (Reportable- Range Exclusions), any change to §2 (Intended Users), §3 (Intended Environment), or §4 (Required Disclaimers) — is a **regulated change**. Required process: 1. Open a change request referencing this document and version. 2. Document the proposed change, the business reason, the regulatory analysis (does it move the product toward SaMD scope? toward LDT scope? toward HIPAA covered-entity scope?), and the validation impact. 3. Obtain written sign-off from the Owner listed in the front matter **and** from a second reviewer from Quality / Regulatory. 4. Bump this document's `Version`, set `Status` to `In Review`, and route for sign-off per `sops/CHANGE_CONTROL.md`. 5. On approval, mark `Status: Approved`, increment `Last Reviewed`, and archive the prior version under `Status: Superseded` for audit history. A change to this document **may** trigger a revalidation of the pipeline (see `technical/PIPELINE_LOCK.md` revalidation triggers) and **may** trigger an update to the customer agreement template (see `compliance/DPA_TEMPLATE.md`; a dedicated `customer/PILOT_AGREEMENT.md` [TBD] ). ## 7. Open Issues That Constrain This Statement These issues are tracked in the `investigations/` bundle and are restated here because they directly constrain what the Platform may claim: - **GIAB v5.0q HG002 30x gap (121,994 missed truth variants) — RESOLVED via reportable-range exclusion.** The gap is now scoped out by the exclusion BED at `investigations/v5_0q_excluded_regions.bed` (per `V5_0Q_GAP_ANALYSIS.md` §5.10, §6 — Outcome 3b firm). The in-scope complement (including the MHC / HLA region per ADR-0006) operates at SNP F1 = 0.9993 / Indel F1 = 0.9959 on HG002 30x (arithmetic estimate; hap.py confirmation pending). Customer written acknowledgement of the exclusion BED is the remaining gate to lift QUALITY_CLAIMS.md C-001 / F-010. - **HaplotypeCaller benchmark output is invalid.** Until fixed or excised, HaplotypeCaller cannot be a quality basis (see `intended-use/QUALITY_CLAIMS.md`). - **GIAB v4.2.1 DeepVariant on HG002 30x is credible for pilot work only.** It is not, on its own, a clinical claim. Phase 4 clinical production requires the customer's own validation per §3.