probiome Sign in
probiome.ai

How we know what we say

Every page on Probiome comes from published research. Most of the work is plain code that gives the same answer every time; probiome.ai’s models read and write; and a person checks before anything is shown.

  • Models read and write

    They read papers into forms and draft summaries. They never decide what’s shown.

  • Code counts and decides

    Rules, counts and the words a page may use. The same answer every time.

  • A person checks

    Every finding before it’s drawn, every summary before it’s published.

Four layers, from the papers up

  1. People

    The Probiome team checks findings, reviews summaries and approves every change to the method.

  2. probiome.ai’s models

    Read papers into forms, answer set questions about them, draft summaries, suggest pairs worth studying.

  3. Code

    Sorts papers by their labels, counts separate groups of people, applies the evidence rules, picks the words. Most of the work happens here.

  4. Sources

    PubMed and Europe PMC, research databases (GMrepo, AGORA2, BacDive), and studies’ own published tables.

What the probiome.ai byline means

Violet always means probiome.ai did something. The word after it says what.

  • probiome.ai · Counted Code counted separate groups of people, from studies’ tables and research databases. Where you’ll see it: “Fewer in 8 of 8 datasets”, on condition pages
  • probiome.ai · Read A model read a study into a form; a Probiome editor checked it. Where you’ll see it: Violet lines with Sources on; probiome.ai’s rows in a panel
  • probiome.ai · Written A model wrote a summary from what code counted; every sentence was checked. Where you’ll see it: Strain and study summaries
  • probiome.ai · Predicted A model suggests a pair no study has tested yet: worth studying, never a finding. Where you’ll see it: Violet dotted rings on the map Not shown yet

Go deeper

Something wrong, unclear or missing? Every page has Report a problem. Each change is dated.
Report a problem →