Fighting Infections with Genomics and AI: Precision Medicine

From PhD to founder

Niamh trained in evolutionary biology and genomics. When she was looking for a postdoc, she was offered a funded position at NYU and, at the same time, a place in a Cornell Tech program that, as she puts it, takes people who just got a PhD and turns them into a CEO. She chose what she calls the weirder, scarier path. She had started small businesses as a child, refurbishing furniture to sell and making perfume from flowers, but she always saw herself as a scientist first.

She spent three years in the program. It opened with a six-week boot camp led by the program’s founder, Uzi De-Haan, which she describes as transformative: thinking about how science can be applied, how to build products, and how intellectual property fits in. What carried her past the uncertainty, she says, was the mission. Cornell Tech helps people build companies that do good, and she had seen how hard funding was for basic science in the US after a series of rejected grant applications.

She and Kevin also see a strong overlap between PhD students and founders. Both have to be persistent and resourceful with little money. As a grad student studying fungi that attack plants, she needed a humidity cabinet that cost $100,000, so she built a tent with reptile-tank bubblers and slept nearby to keep the humidity steady for days. Kevin says that when he reviews PhD candidates, evidence of bringing an idea to fruition often matters more to him than their niche.

Precision medicine for infections

Biotia’s mission, Niamh says, is to bring precision medicine to infectious disease diagnostics, with many of the gaps in women’s health. The standard of care is to take a sample, such as urine, grow it on a petri dish, look under a microscope, and identify the pathogen, a method she notes dates back to the 1800s. She and her co-founder, Chris Mason of Weill Cornell Medicine, used genomic tools every day in research and were frustrated that the clinic didn’t. Niamh had her own experience: a hospital-acquired urinary tract infection while she was pregnant.

Kevin describes what a failed culture means for a patient: a serious infection, sometimes sepsis, with no clear cause, so doctors treat broadly and hope the patient can tolerate it. Biotia’s first test, the BIOTIA-ID Urine Test, is built for people with chronic or complicated urinary tract infections, who by her description get infected at least two or three times a year and are often treated empirically, sometimes for months. She says about a third of sepsis cases begin in the urogenital tract, and that the test was built for an estimated 10 million US patients with chronic UTIs or interstitial cystitis.

Sequencing avoids the need to grow anything. Every microorganism has DNA or RNA, Niamh explains, so it can be extracted and sequenced directly, including species that need special conditions, grow slowly, or don’t grow in standard culture at all.

Building and validating a clinical lab

Turning research into a test started with research and development and a lab of their own. Biotia runs a clinical lab in New York City, which needed federal CLIA certification and approval from New York State’s Clinical Laboratory Evaluation Program, a standard Niamh calls the highest bar they are held to. State inspectors audit the lab’s methods and staff qualifications. The lab now uses robots to automate much of the genomics workflow.

Getting a reliable signal every time took work that a research study doesn’t require, she says: different sample volumes from men and women, weak early results from men’s samples, and sample stability. After a full validation, the test was under New York State review for 18 months. She attributes part of the delay to the test’s AI component, which arrived during the pandemic while regulators were still hiring AI experts and working out how to review this kind of product. Surviving that gap as a startup, she says, was very challenging.

Machine learning that regulators can trust

Kevin outlines the assay as metagenomic sequencing: sequence the DNA and RNA in a sample, set aside the human DNA, and analyze the rest. Niamh says one key use of AI is a machine learning classifier that separates species that look nearly identical. Her example is E. coli and Shigella, which look alike genetically and under a microscope. Biotia grew E. coli, gathered Shigella genomes, and trained its classifier on the difference, then did the same for strains, antimicrobial resistance markers, and virulence markers.

The team now predicts a resistance phenotype from the genotype, not just the presence of a resistance gene. Niamh says Biotia won the CAMDA international competition on this task two years in a row. Kevin notes this is not the generative AI people usually picture. Niamh agrees: it’s “not the scary AI.” With models such as random forests, the same input gives the same output, which she says makes reviewers of regulated diagnostics more comfortable. Kevin adds that probabilistic language models are what make AI worrying in regulated settings.

The long road to insurance coverage

Before building the assay, Biotia did market research with clinicians and patients. Selling it, Niamh says, depends on reimbursement, a long process the company is in the middle of. The urine test has a Proprietary Laboratory Analyses (PLA) code, a test-specific billing code, and she says it recently received a final price determination. The next step is submitting claims to Medicare one at a time, each argued with doctors’ notes, and many will be rejected. After that, Biotia plans to submit its clinical studies on outcomes and cost to seek a local coverage determination, which would mean payment without arguing every claim.

Payers’ standards are very high, she says, and they want evidence in specific populations: kidney transplant recipients, people with neurogenic bladder, cancer patients, and others who are immunocompromised, for whom a urinary infection can become sepsis more easily and who are more likely to carry drug-resistant or atypical pathogens. Coverage is carved out one group at a time. Biotia is also developing a test for prosthetic joint infections with Hospital for Special Surgery. These infections, after a hip or knee replacement, are hard to grow from viscous joint fluid, and a wrong call can lead to revision surgery.

Asked to steelman the insurers, Niamh points to Thomas Kuhn’s account of scientific revolutions: science and medicine move slowly, and slow adoption in hospitals can protect patients from treatments that aren’t yet proven. She calls the US insurance system broken, but says payers can’t fund every expensive test, and that Biotia has worked to bring its price in line with a PCR panel.

Clinical proof from culture-negative patients

Niamh describes a retrospective study with SUNY Downstate in Brooklyn of 200 patients being treated for urinary tract infections whose cultures had grown nothing. Biotia’s test found urogenital pathogens and drug resistance markers in 63% of them, she says. Comparing the prescriptions with the results, some patients had been given the wrong antimicrobials, others could have had treatment narrowed, and a group of catheterized patients with fungal infections had been treated with antibiotics. A prospective, interventional study, in which clinicians treat based on the test and outcomes are tracked, is being set up.

Biosurveillance: who pays for it?

Biotia began with environmental testing in hospitals, looking for pathogens such as C. diff on surfaces and equipment, and with biosurveillance: global microbial maps, drug resistance, and models of how climate could affect diseases such as dengue. The problem, Niamh says, is who pays. She believes governments should fund surveillance continuously, but money tends to arrive during a pandemic and lapse afterwards. Diagnostics was the clearer business because it is obvious who pays: the patient, the practice, or the insurer.

Kevin explains that hospital surveillance usually falls to infection prevention teams with small budgets. Niamh recalls a study, she thinks from UPMC, showing how siloed hospitals are: the department that would pay for surveillance isn’t the one that bears the cost of an outbreak. Biotia could sell surveillance to a hospital during an outbreak, she says, but those are rare events people try to forget. It now has some hospital contracts, including isolate work with Mayo Clinic, and pools the data from its tests into a biosurveillance platform for population-level insight.

One lasting gain from the pandemic, in her view, is that wastewater testing continues. Everyone flushes, so a treatment plant collects signals from a whole population in one place, and sequencing or PCR can track trends in infections such as COVID-19, flu, and RSV. Kevin notes that the same metagenomic approach underlies Biotia’s urine test.

The moat, and patent then publish

If the underlying technology is shared, Kevin asks, what is the moat? Niamh points to the validation, to large data sets curated with AI and by experts, and to patents: two issued software patents, which she says are not the strongest kind, and a pending one that combines wet-lab methods and software for handling tricky samples and inhibitors. Coming from academia, the team patents defensively and then publishes, rather than keeping methods as trade secrets.

Kevin mentions others bringing metagenomic tests to the clinic, including Delve Bio and the work of Dr. Charles Chiu at UCSF. Niamh says her co-founder has published with Chiu, who advises Biotia on space genomics: on Earth they are competitors, in space they are collaborators.

Microbes in orbit and diagnostics for remote clinics

Biotia has worked with NASA on samples from the International Space Station. Niamh says the team identified novel fungi and bacteria, likely brought aboard by astronauts, that colonize the station and appear to be evolving under the different pressures of space. Chris Mason has also studied astronauts who spent long periods in space, including changes in gene expression, the microbiome, and telomere length.

With Dr. Charles Chiu, Biotia was funded by the Korean pharmaceutical company Boryung to develop a test that sequences and analyzes samples during spaceflight, using a palm-sized Oxford Nanopore MinION. The goal is astronaut health, she says: spaceflight can dampen the immune system, latent pathogens can resurface, and only a limited set of antimicrobials is on board. She notes that her co-founder has written about the next 500 years of humanity in space.

The same tools could help on Earth, Niamh argues. An unbiased test that can find anything, rather than a panel for specific pathogens, suits remote clinics facing outbreaks such as Ebola. Biotia recently published a urine study with Bumrungrad International Hospital in Thailand using nanopore sequencing, where uploading data was sometimes a problem. Today, she says, a single sample can be sequenced and analyzed on a laptop, though not the thousands Biotia runs in its lab. Kevin adds that running compute on site also takes technical expertise that is hard to support remotely.

Taking Biotia international

Part of Biotia’s team, including its chief medical officer, is in Budapest, and the company has looked at labs there as a possible entry into Europe. It also works with partners in Southeast and East Asia. Because the team is still small, Niamh says, it is focusing on US regulation and reimbursement first. Many patients abroad ask to pay for the urine test themselves, so the first step will probably be offering it in Europe and Canada on a cash-pay basis.