Susceptible Infections that Do Not Stay Susceptible
When a hospitalized patient shows signs of a bacterial infection, clinicians often begin treating it, even if they do not yet know which microbe is responsible. It then takes two to three days for the laboratory to identify the organism, test its antibiotic susceptibilities and adjust the treatment if appropriate. That process protects the patient by starting treatment immediately, while gathering evidence to select the best antibiotic. But it’s only a snapshot, and pathogenic bacteria often evolve resistance after treatment begins.
“All of the testing that we do right now is a static point in time, and it’s really only looking at that bacterium in the moment,” says Kalen Hall, cofounder and chief executive officer of Informuta, a Tulane University spinout developing genomic tools to predict treatment-emergent resistance.
Antibiotic treatment can drive an infection to become drug-resistant in two ways—by selecting for an existing subpopulation of resistant bacteria or by selecting for a new resistance mutation. In either case, the resistant organisms multiply, and a patient who initially improves may deteriorate after the therapy loses effectiveness.
This phenomenon, known as treatment-emergent resistance, is difficult to quantify because it requires data on isolates collected at successive time points, as well as on treatment and outcome. In unpublished data, Informuta has observed it in 10–40 percent of Pseudomonas aeruginosa infections, including lung infections such as cystic fibrosis, as well as other respiratory infections and urinary tract infections. The rate depends on the antibiotic used and the clinical context, although Hall says its overall prevalence remains uncertain.
Reading the Genomic Signature
Informuta grew from Hall’s doctoral research on hypermutating P. aeruginosa in individuals with cystic fibrosis. These bacteria can lose the ability to carry out a process called mismatch repair that fixes mistakes that arise when a bacterium duplicates its DNA as it prepares to divide into two new cells. This causes the daughter cells to accumulate mutations rapidly, including those that cause multidrug resistance.
In 2025, Hall and colleagues reported in Nature Communications1 that they had identified a characteristic mutational signature in P. aeruginosa strains deficient in mismatch repair. In laboratory experiments, these mutation-prone bacteria acquired resistance rapidly under repeated antibiotic exposure. They also developed cross-resistance to other drugs when resistance to the initial and secondary drugs could arise through shared mechanisms.
The researchers then analyzed clinical isolates, and the results showed that when the mutational signature they had discovered was present, the pathogenic bacteria were both deficient in mismatch repair and more likely to have developed multidrug resistance. The same isolates developed resistance rapidly when Hall and her colleagues tested them in the lab.
The study provided a proof of principle, but its scope was limited. It focused on a single pathogen, a specific DNA repair defect, selected drugs, and experimental conditions that cannot reproduce the full complexity of a patient infection.
Informuta says it now combines multiple genomic signatures with longitudinal phenotypic data to train machine-learning models for specific pathogen–drug combinations. Rather than searching only for known resistance genes or individual mutations, Hall and her colleagues adapted an approach from cancer genomics. The company’s platform examines mutational signatures—genome-wide patterns that reflect the biological processes shaping a bacterial population.
“We don’t really care what’s changing; we just care about the pattern of the changes,” Hall says.
The aim is not simply to label a bacterium as likely to mutate, but also to estimate whether resistance is more likely to emerge against one antibiotic than against another.
Informuta is currently testing whether their approach can be extended to other bacterial pathogens. Each species has different genomic characteristics and requires separate calibration, Hall says.
From Prediction to Clinical Action
Informuta expects its test to supplement rather than replace conventional antimicrobial susceptibility testing. Current methods, including broth microdilution and disk diffusion, determine which antibiotics can stop or kill the bacteria. In contrast, Informuta’s approach would estimate the future risk of resistance.
Such a forecast might support selection of another antibiotic, closer monitoring, earlier repeat testing, or specialist review.
Before Informuta’s method is used clinically to select antibiotic treatments, it will be used for risk stratification, meaning that it would estimate the risk of drug resistance developing during antibiotic treatment. The method is probabilistic, meaning that a low-risk result would not guarantee that resistance could not emerge, and a high-risk result would not mean that it inevitably would.
Errors in either direction carry consequences. A false negative could leave a patient on a drug that soon becomes ineffective. A false positive could prompt unnecessary treatment escalation, potentially conflicting with antibiotic stewardship principles, which aim to minimize unnecessary antibiotic use to slow the development of resistance.
Next up for Informuta, Hall says, is to compare its model’s predictions with data obtained by tracking isolates from patients over the course of an infection and testing them for antibiotic susceptibility. The company is also preparing its laboratory to test patient samples and working with health-system partners to do clinical pilot tests.
“Ultimately, the duty is on us to prove that it’s useful,” Hall says.
That will require more than just showing that the model predicts whether bacteria resist higher antibiotic doses in a petri dish. Informuta must also demonstrate that when clinicians act on the information, it improves outcomes without promoting unnecessary antibiotic escalation.
The Clinical and Economic Test
Even if Informuta’s forecast is accurate, it could still fail if it arrives too late, is difficult to interpret, or does not generalize across hospitals. Ultimately, it must work under the imperfect conditions of clinical care.
Adoption will also depend on economics. Hospitals may recognize the public-health value of slowing antimicrobial resistance, but broad preventive benefits do not necessarily justify a purchase when budgets are constrained. Informuta will need studies to measure more immediate effects, such as treatment failures, length of stay, readmissions, intensive care use, rescue interventions, and total treatment costs.
Informuta plans to focus initially on high-risk oncology and transplant infections, where treatment failure can be especially consequential.
If the first application succeeds, potential future applications of Informuta’s platform could include supporting antibiotic development and research on treatments that combine antibiotics and hospital-level surveillance. For now, its future rests on a narrower test: whether genome-wide patterns can predict clinically meaningful resistance accurately enough—and early enough—to change the treatment of a patient whose infection still appears susceptible.
Even in the best-case scenario, predicting antimicrobial resistance does not mean eliminating it. But it could allow early susceptibility testing to address a question it cannot answer today: not only which antibiotic works now but which is most likely to keep working.
