New antibiotics can't win the economics. So we don't build them. Subsira's AI reads a pathogen's genome and finds what makes each antibiotic it's resistant to work again — in minutes.
The partner for a failing antibiotic already exists. The bacterial genome selects the right one — until now, this has been done empirically. We turned that into an inference problem: genome in, matched partner out, in minutes.
Why the pairing holds: a resistant strain can defend itself against the antibiotic, or against the agent we pair with it — and those two defenses trade off against each other. Which trade-off applies to this strain is written in its genome.
The matched partner — what we mean by it A second agent, selected by the pathogen's own genome, paired with the antibiotic that strain has defeated — chosen because escaping it costs the strain the defense it was using against the antibiotic. Today that agent is a matched phage. Next it's a matched β-lactamase inhibitor. It is never a generic potentiator handed to every patient — if the genome doesn't choose it, we don't ship it.
You can't out-invent a market this broken. So we don't — we match the partner the genome already points to.
The engine is the durable asset; each modality is an instance of it at a different genomic resolution. We add a modality only where a feature in the genome names the agent.
Matched down to the individual isolate, on the receptors and defenses carried in its genome. No two strains get the same answer.
The resistance genotype names the enzyme outright, and the enzyme names what restores the β-lactam.
Capsule, surface and defense loci are legible in the same sequence, and each one names an agent of its own.
Same engine, finer resolution each time — strain, then enzyme, then locus. The genome selects the agent at whichever resolution the mechanism lives. We never add genome-independent potentiators.
From restoring one failing antibiotic to the decision layer for every resistant infection.