In the last decade, “personalization” has become one of the most overused words in medicine. But even with the many promises to tailor care for each patient, millions still shuffle through the same diagnostic routines, prescriptions and side effects.
It’s the paradox shaping much of AI-powered healthcare today. While artificial intelligence has made some major parts of healthcare more efficient, it’s not necessarily made it more human. It’s like an illusion of individual care that’s left many patients feeling more like data points than people.
The problem, experts say, isn’t the technology itself but the kind of data we feed it. Most AI models in healthcare today are trained on population-level datasets such as electronic health records and claims information. While those datasets can reveal statistical trends across millions of patients, they rarely capture what’s really happening inside any one of their bodies. In other words, they can predict probabilities, but not biological realities.
As Mika Newton, CEO of xCures, said earlier this year, “AI will not transform healthcare if it operates in a vacuum. AI requires the foundation of high-quality data, which begins with patients. That personalization problem is what startups like California-based Parallel Health are now trying to solve by helping AI interpret biological data directly.
From Data Points To Living Systems
“Real personalization means treating you as a complex system, not a statistic,” said Natalise Kalea Robinson, cofounder and CEO of Parallel Health. “Most ‘personalized’ healthcare today is really just sophisticated segmentation — you’re placed in a bucket based on symptoms — sometimes demographics or genetic markers (if you’re lucky), then given the treatment that worked for most people in that bucket.”
Parallel’s work is one example of how companies are beginning to use biology as the foundation for personalization rather than relying only on medical records or demographics. Its platform uses quantitative whole-genome sequencing to map the trillions of bacteria, viruses, and fungi that make up a person’s skin microbiome. “This isn’t about comparing you to a population average — it’s about understanding your individual biological reality at the microbial level and at multiple points,” Robinson explained.
Two patients may share an acne diagnosis, but their underlying causes can be entirely different. One patient might have an overgrowth of Cutibacterium acnes phylotype 1A, the bacteria usually linked to acne, while another might have antibiotic-resistant types that explain why regular treatments didn’t work. “No two ‘acne’ patients have the same skin microbiome; we have yet to see that across our incredibly large data set,” Robinson said.
That biological specificity allows the company to design targeted phage serums that eliminate only harmful strains while preserving beneficial microbes. Outside researchers agree that this type of precision is scientifically promising, though they note that phage therapy still faces steep regulatory, manufacturing and standardization hurdles before widespread adoption.
Robinson acknowledges those challenges but argues that adaptability — not just precision — will determine which approaches endure. “Your biology isn’t static, so your treatment shouldn’t be either,” she said. “Real personalization is longitudinal, adaptive, and grounded in your actual biological data — not population proxies.”
Teaching AI To Understand Cause And Effect
For years, healthcare AI has been praised for pattern recognition — spotting tumors in scans, predicting readmissions and flagging anomalies in lab results. But Dr. Nathan Brown, Parallel Health’s chief science officer, argues that’s only the surface. “Working with direct biological data transforms AI from a pattern-matching tool into a mechanistic prediction engine,” he said.
By analyzing how microbes interact with one another and with the human host, the system can begin to infer causality rather than mere correlation. “Our AI can identify that specific microbial imbalances preceded symptom onset by months, enabling true prediction, not just early detection,” Brown noted.
That insight, he said, turns AI from reactive to preventive medicine. The same microbial patterns that signal inflammation in acne, for instance, may also appear in conditions like rosacea or certain types of psoriasis. “What we learn about microbial dysbiosis in one condition can apply to others. Our AI is learning fundamental principles of host-microbe interaction that generalize across diseases. We then have the power to redefine complex diseases.”
While independent researchers have echoed the potential of biology-driven AI systems, especially those based on microbiome data, they remain cautious, as noted in a review published in Nature.
Scaling The Science
The word “personalized” often evokes hand-crafted medicine — treatments so specific they can’t possibly scale. Dr. Seaver Soon, Parallel’s lead dermatologist and clinical advisor, said that assumption misses how platform technologies evolve.
“Personalization doesn’t mean we’re creating unique treatments for every individual from scratch,” he said. “We’re using platform technology to efficiently match patients to a bespoke solution from a defined toolkit.” Parallel claims its ‘toolkit’ draws on an expanding biobank of microbial strains and a manufacturing process aimed at stabilizing targeted phage therapies — a challenge the broader biomanufacturing field is also racing to solve.
That model mirrors the early days of genomic medicine, when sequencing DNA was slow and expensive but eventually became routine. The same could happen with microbiome-based care as the technology matures. “Precision medicine eliminates trial and error,” Robinson explained. “If we can tell from the start that a patient’s bacteria are resistant to certain antibiotics, we can avoid treatments that won’t work, saving both time and cost.”
Recent research supports that idea, with a review in the Journal of Translational Medicine noting that while precision therapies can improve outcomes and reduce waste, their cost-effectiveness still depends on reimbursement policies and access — two long-standing barriers to progress in clinical genomics.
The Ethical Edge
As biology-driven AI becomes more powerful, questions regarding privacy and equity are becoming increasingly prominent. A report from the National Center for Biotechnology Information warned that “the use of large datasets in AI systems has led to discussions about ownership and management of data,” adding that data sovereignty — the right of individuals or groups to control how their biological data is collected and interpreted — will define the next phase of health innovation.
According to Robinson, that principle is already built into Parallel’s model. “Patients must know what data is collected, how it will be used and what they get in return,” she said. “Just because you can collect biological data doesn’t mean you should.”
She believes that transparency and equitable access must coexist. “The most dangerous risk in personalized medicine is creating a two-tier system where precision care is available only to the wealthy. “Communities that contribute data to our AI models must benefit from the resulting improvements.”
Bioethicists are increasingly voicing similar concerns. Recent research — including a 2024 paper in BMC Medical Ethics by Shaw and colleagues and a 2025 study published by the Committee on Data for Science and Technology — emphasizes that the future of personalized medicine will depend not only on smarter algorithms but on fairer systems of trust, consent and shared benefit.
Robinson calls it a shift of power back to the patient — a much-needed correction at a time when data privacy remains a defining issue. Whether healthcare follows that path will depend on AI’s ability to account for the biological complexity of each individual, rather than just patterns in population data.