7448 AI · Simulator
Change the bacteria, watch the chemistry move.
Every slider starts at the average level found in healthy people. Move one, and the model re-predicts the whole chemical profile. Everything you are not adjusting stays at its real healthy-population value, because a profile of five known bacteria and 5,033 zeros is not a gut that exists.
Loading the model…
Or start from a real person's sample
Published, de-identified research samples from the 14 source cohorts.
Sliders apply on top of this real sample rather than the healthy average.
Predicted chemical output
The number beside each metabolite is the change from the starting profile. Changes are more reliable than absolute levels, because the model's systematic error largely cancels in a difference.
Literature context
What the literature says about profiles like this.
These are not predictions about you or anyone. Where a bacterium in this profile sits well away from the healthy average, we look up published studies that reported the same shift in the same direction in a given condition, and show you those studies.
Population
Where this profile sits among real people.
Every dot is one real stool sample from the source cohorts. The marker updates live as you move the sliders.
Each dot is positioned by overall microbial composition (principal components 1 and 2). These two axes capture28% of the total variation, so being near a cluster is suggestive, not conclusive. Cases and controls overlap heavily, which is the honest picture of how much a microbiome profile alone distinguishes them.
Questions
A result you want to talk through?
If a prediction surprises you, or you want to run 7448 AI on your own sequencing data, write to the team that built it.