GeneMachine / evaluation / October 2, 2026
A useful bridge from evidence to better questions.
The strongest opportunity is an understandable, private experience that helps people see what genetics can contribute to a medicine discussion and what their data cannot establish.
The judgment
Promising product direction. Breakthrough unproven.
Pharmacogenetic interpretation already has established tools and guidelines. GeneMachine should compete on comprehension, access, provenance, and honest handling of unknowns. A clearer interface and private processing are useful product improvements; they do not demonstrate scientific novelty or improved health outcomes.
PharmCAT already handles broader genomic inputs and does not claim useful comprehensive interpretation of consumer arrays. It also operates locally. Privacy alone is therefore insufficient differentiation. 23andMe already offers a limited set of pharmacogenetic reports.
Useful today
Three audiences, concrete outcomes.
People learning about a medicine
Search seven gene–medicine topics or write a question about any medicine. Edit the wording, add notes, and put your main concern first. No kit, account, or genetic disclosure is needed.
People with consumer DNA
Inspect file quality, then see one gated SLCO1B1 observation or an explicit reason it cannot be reported. Cancel analysis, revoke consent, or clear the result.
People preparing for care
Copy or download a worksheet with your questions and notes, clearly separated from sourced library context. A medicine absent from the library is marked as unassessed. A separate DNA discussion report preserves its narrow observation and limits.
The current product is an English-language educational app. It does not meet every person’s clinical, language, accessibility, or data needs. Broad usefulness must be tested with real users rather than assumed from the interface.
A boundary the design must preserve
Seven learning topics. One DNA observation.
| Surface | What it can establish | What remains unavailable |
|---|---|---|
| Medicine library | General context from published CPIC guidance and useful questions. | Individual response, a recommendation to test, dose, or treatment choice. |
| Consumer DNA checker | File-level consistency for rs4149056 when build, coordinate, strand, allele, and uniqueness agree. | Clinical accuracy, phenotype, star alleles, CYP2D6/HLA, phase, copy number, or comprehensive PGx. |
| Reports | Your questions and notes with curated context, or a derived DNA observation, limits, and cited sources. | A clinical laboratory report or an automatic medical decision. |
The library does not match its topics to DNA. Missing markers stay unknown. The FDA association table does not imply that everyone should be tested. The CPIC statin guideline provides clinical context; it does not validate GeneMachine.
What could earn a breakthrough claim
Prove that people understand and use the evidence.
Comprehension and access
Test whether first-time users complete the checklist, understand that unknown is not normal, distinguish general evidence from personal findings, and know when clinical confirmation is needed. Compare the same tasks with their current workflow.
Then prioritize translation, assistive-technology testing, and the medicines users actually need. New topics require primary evidence and qualified review.
A dependable clinical handoff
A future validated genomic lane could expose callable and missing sites, tool/evidence versions, and reproducible reference checks before producing a report. PharmCAT offers an existing foundation to evaluate for suitable VCF inputs.
Require reference-material validation, a qualified pharmacogenomics review, and evidence of clinical utility before personalized medication claims. Do not fill sparse-array gaps with AI guesses.
Delivery and remaining evidence
Software checks support an educational candidate.
The repository contains unit tests and actual Chromium/WebKit journeys for search, custom and edited questions, ordering, copying and export, technical abstention, local processing, cancellation, consent withdrawal, reset, and narrow layouts. The static build contains only public app assets and a synthetic example. The delivery receipt records the actual commands and results.
Clinical validation, qualified clinical content review, real-user comprehension studies, and native iPhone testing remain unperformed. A static build is separate from public deployment. Clinical use is outside this candidate’s supported scope.
Sources and update responsibilities
The appointment workflow follows AHRQ guidance on preparing medicine questions. MedlinePlus explains that pharmacogenetic tests are not available for every medicine. Custom questions are not clinical assessments or evidence that a test is appropriate.
Source URLs are pinned in lib/medicine-library.mjs and the source register. Static content does not update itself. CPIC’s July 2026 DPYD notice is carried into the topic and checklist. Review that topic when the new guideline is published.
Provider download instructions link to current official 23andMe and Ancestry pages. No current kit price or lowest-cost claim is made.