Analysing genomic data without giving it away

A whole genome is about 100 gigabytes of raw data. It contains everything: disease risks, ancestry, relatedness, traits you may not even know about yet. Anyone who analyses this kind of data faces the same quiet decision almost every time, usually without making it consciously: do you upload it to a cloud, or not?

The convenient answer is the cloud. Most analysis services are web portals: upload a file, wait, download a report. That is easy, but it means the most intimate data a person owns sits on someone else’s server: copyable, indexable, bound to a jurisdiction that may not be your own. For a clinic, a lab, or a researcher, that is often not just uncomfortable but simply not permissible under regulation.

Genome takes the other path: the analysis runs entirely locally on your Mac. No upload, no telemetry, no account. The data never leaves the machine. This piece explains what that means technically and why “local” is more than a privacy slogan.

The pipeline stays on the device

A genome analysis is not a single operation but a chain: raw data (FASTQ) is aligned against a reference genome (BAM/CRAM), from which specialised analyses are run (HLA typing, KIR genes, pharmacogenetics, repeat expansions), and at the end there is a report.

In a cloud solution, a copy of the data travels across the network at every one of these steps. In Genome, the whole chain runs inside a native macOS application. The established bioinformatics tools (HLA*LA for HLA typing, T1K for KIR, Aldy for pharmacogenetics, ExpansionHunter for repeat regions) are orchestrated and managed locally. You can see which tools are installed, in which version, from which source. None of it is a black box in a data centre.

This has a pleasant side effect beyond privacy: reproducibility. When the toolchain is visible and versioned, a result can be traced later. That is a property you should expect from software that prepares medically adjacent statements.

A report that carries its own origin

The most valuable part of an analysis is not the number, but the trust in the number. Genome produces PDF-first reports that show not only results but the data basis, the evidence, and the limits alongside them: what does the statement rest on? How good was the coverage at this position? What can this test not say?

Each report also carries a provenance record: among other things, a SHA-256 fingerprint and the app version it was generated with. That sounds technical, but it is the essence of seriousness in genomics: a report that is reproducibly traceable to a specific input and a specific software version, rather than a pretty but unverifiable graphic.

Why “local” is a feature, not a compromise

There is a widespread misconception that local software is the weaker, compromised variant of the cloud. For sensitive health data it is the other way around:

• Privacy by design. What never leaves the machine cannot be leaked, resold, or seized at a third party by court order. GDPR compliance shifts from an ongoing task to a property of the architecture.
• No dependency. No service that can be shut down, no API that changes, no price increase that makes access to your own data more expensive.
• Full control over the toolchain. Anyone who needs to know exactly which version of which tool produced a result (a must in research and clinical settings) only gets that if the tools run locally and inspectably.

The price is compute on your own device and an application that translates the complexity of a bioinformatics pipeline into a calm, native interface instead of hiding it behind a web form. That is precisely the task we set ourselves.

Who this is for

Genome is for people who analyse genetic data and want to keep control while doing it: researchers, small labs, technically minded users who want to understand their own raw data, and anyone for whom “the data sits on our server” is not an acceptable answer.

If that sounds like your problem: Genome runs on macOS, processes FASTQ, BAM and CRAM locally, and the first report is produced without a single byte leaving the machine.