Food genomic material was detected only in about half of infant stool samples but increased at the onset of solid food consumption and was ubiquitous in adult stool samples.

We also ran a proof-of-concept identifying foods and nutrients that were associated with the onset of metabolic disease in an adult cohort.

Building a comprehensive genomic database for as many foods in FOODB as possible we could connect individual genomes to nutrient content. For now we can match 77% of all foods in FOODB with taxonomic information and the next version of the database will push this to 90%.

Using a decoy-aware mapping approach with additional consistency filtering we could show good sensitivity and specificity in simulated sequencing samples with a false positive rate around 1-10 reads per million.

Dynamic models don't have that issue because they don't maximize community growth, but they are a bit costly to simulate, especially if we only want to predict steady state fluxes.

How do you get around that? Use a steady state method that predicts growth rates that look like they could have come from a dynamic model, for instance by enriching for growth rates that can be reached easily from inoculation.