Bulka, Ghassemi Nedjad, and Pauleve (2602.21993) predict what microbes eat from their metabolic blueprints — solving a problem that currently requires expensive trial-and-error cultivation.
Metagenomics can identify novel microbes by sequencing environmental DNA, but studying these organisms deeply still requires growing them in culture. The bottleneck: you need to know what nutrients a microbe requires, and for uncultured species, you don't know this. Current practice involves testing many growth media combinations, which is slow, expensive, and often fails because the correct nutrient combination was never tried.
The approach reverses the logic. Instead of testing nutrients empirically, it reconstructs the metabolic network from genomic data — the set of biochemical reactions the organism can perform — and then infers which external nutrients are necessary. If the network contains all the enzymes to synthesize amino acid X from simpler precursors, the organism doesn't need X supplied externally. If it lacks a critical step in X's biosynthesis pathway, X is a source nutrient: it must come from the environment.
The subtlety is in network gaps versus genuine requirements. A missing gene might mean the organism truly can't synthesize a compound, or it might mean the gene exists but wasn't identified in the metagenome. The prediction algorithm has to distinguish between biological absence (a real nutrient requirement) and annotation absence (a database gap). Getting this wrong in either direction wastes resources: predicting a nutrient as required when it's not adds unnecessary cost to the growth medium; predicting it as synthesizable when it's required means the culture fails.
The value is practical: turn a genomic sequence into a recipe for growth medium, before ever touching a petri dish.