lucashuizinga/dnadistiller

DNAdistiller takes a raw DNA export from services like 23andMe or AncestryDNA and turns it into a short, focused report about aging and long term health, instead of handing over the full raw file. A typical export contains around 640,000 genetic variants. This tool reduces that down to a small curated set, twelve markers in its first version, chosen because each one plausibly relates to how long someone stays healthy rather than to unrelated traits like earwax type or eye color. The tool runs entirely on the user's own computer. There is no account to create, no upload step, and no network activity involved in generating a report. Once installed, a person points it at their downloaded DNA file and gets back a plain text profile organized into categories like lipids, metabolic risk, and drug response, with each finding rated by how strong the supporting evidence is. Markers with weak evidence are hidden by default unless specifically requested. A marker related to cognitive aging risk is also hidden by default and only shown if the user opts in, since that is information some people would rather choose to see rather than encounter unexpectedly. Each finding in the report is written so it can be read in under a minute, and it can then be pasted into an AI chat assistant like Claude or ChatGPT to discuss further, without ever sharing the rest of the person's genetic data. The report also points out which findings are better confirmed with an actual blood test rather than trusted on genetics alone, since a blood test often gives a more direct answer than a genetic prediction can, and it separates markers into those a test can replace, those worth checking alongside, and the rare few where no test warns you in advance. The project is written in Python and requires Python 3.10 or newer. It can be installed using the uv tool, or built from source by cloning the repository and installing it with uv as well. It supports raw files from 23andMe, AncestryDNA, MyHeritage, FamilyTreeDNA, Living DNA, and single sample VCF exports. The project is still early, at version 0.1, with a small and carefully chosen catalog of markers rather than a long list of speculative ones.