As an associate professor of medicine at the University of Chicago, Pearson directs the Center for Computational Medicine and Clinical Artificial Intelligence. He also practices in the Section of Hematology/Oncology, effectively putting two feet into different worlds he is trying to unify. One is in the world of the clinician, where the decisions get made. The other is the computational side, where the work, according to Pearson, should include ensuring the clinical side is better informed. The challenges are not small, but the rewards of solving them are profound.

The information gap between these two worlds is not simply a research abstraction. Biomarkers, the measurable molecular signals that indicate which treatment a given tumor is likely to respond to, are imperfect for most cancers. More than 26,000 U.S. adults diagnosed with advanced stages of cancer showed that only about one third of cancer patients received molecular testing before their first line of therapy, according to a 2025 JAMA Network Open study. While that number is rising – the same study showed that the numbers jumped from 32 percent in 2018 to 39 percent in 2021 and 2022- the bottom line is that tumor-specific information is reaching treatment planning unevenly, including by cancer type, such as where 17 percent in ovarian cancer to 45 percent in non-small cell lung cancer was reported. While the finding does not mean that every patient should receive every available test, it shows that consistency is not currently built into cancer treatment diagnosis.

The problem is not a lack of raw data, says Pearson, but that the underlying data are becoming harder to interpret as more becomes available. He describes the challenge as “maybe not the total quantity of data so much as the complexity of the data.” Cancer screenings now generate many types of information at once. A single case can include pathology images, CT (computed tomography) scans, mutations, gene expressions, and other biological readouts for clinicians to decode as they try to determine the right course of treatment. Building a multimodal biomarker means combining the right data types while ignoring the wrong ones. The difficulty is finding which features, among thousands of possible variables, carry a reliable, clinically meaningful signal.

For Pearson, the route into this problem began with fundamental data that every cancer center already relies on: the pathology slide. This familiar glass slide that practically every high school student has swabbed their cheek for in science class to examine under a microscope is, as Pearson puts it, “ubiquitously everywhere in the world for very low cost.” Early in his career, as deep learning for image analysis became practical, Pearson and his colleagues proposed that the visual patterns in those slides, the way cells are arranged and grow, might carry information about a tumor’s molecular state. The premise was that if computational models could reliably detect these patterns from digitized images of these slides, an inexpensive image could stand in for testing that would otherwise cost thousands of dollars. If so, a computation could use data that are already commonly collected in cancer diagnosis and treatment rather than requiring additional (and expensive) data collection approaches. This shifts the focus from what additional data must be collected to what more can be learned from data already at hand.

This is where high-performance computing (HPC) comes in, Pearson says. A supercomputer’s ability to process large-scale computation gives researchers an expanded toolkit that lets them train, test, and repeatedly evaluate computational models created from large image collections. From there, they can determine if inexpensive, abundant inputs (for example, cells from a pathology slide) can deliver the same information that once required more costly data collection and processing. 

“How much information can truly be extracted from existing data infrastructures with the assistance of high-performance computing, effectively getting more out of the same amount of information,” Pearson posed. 

The same infrastructure supports work on model transparency, uncertainty and reliability, he explains, adding that these capabilities matter because a statistically impressive result is not enough for clinical use.

“A model must produce information that can be validated, understood within it limits and integrated into a decision process,” Pearson says. “And errors ultimately have consequences.”

The potential for very real consequences keeps Pearson grounded. Despite the success of his early modeling, Pearson is not predicting an abrupt transfer of medical judgment to machines. He is circumspect about adoption.

“Medicine is a discipline that is all about incremental, incremental, incremental gains,” he says. He does not envision “one model to rule all of medicine that arises spontaneously in the next six months.”

What he does see, however, are ways to bring the technology into practice incrementally. Measured improvements that can be evaluated one task at a time. Pearson says one such enhancement is already changing practice is an artificial intelligence (AI)-enabled scribe that helps him “talk fluently with the patient in the room, face to face, instead of facing away and typing on a keyboard” during consultation. By transcribing the discussion as it happens, the tool improves one task without taking on any clinical judgment, which is exactly the shape Pearson expects adoption to take: specific things done better, one at a time.

In this model, Pearson is clear that AI enters clinical care as cautious physicians take up the tools only if they feel that “at every stage we’re moving in the right direction, rather than ceding all of medicine to an AI companion or assistant.” Much of his own work on these systems is methodically improving the transparency and reliability of models before routine deployment, building trust and validating in the areas where it performs well, learning where it does not and carefully learning how its outputs should inform care. Trust develops through evidence and controlled use rather than through scaling the technology alone.

Pearson notes that the barrier of entry for using these sophisticated models is falling fast. Low-code interfaces and agentic software-development tools have swiftly reduced the coding expertise a clinical researcher needs to use a sophisticated model. The result is growing interest in using HPC as researchers better realize its power to help them and their patients. At his own institution, the University of Chicago, he says, the past six months brought “a substantial increase in use of the local high performance computing infrastructure,” as more researchers could actually understand and access its capabilities in a way that augments their practice.

For Pearson, this ability to use these technological tools has brought him to the doorstep of quantum computing, which sits at the far edge of this expanding computing infrastructure. Pearson is careful to frame it as a bounded research question, noting that the appeal of a quantum subroutine was combinatorial optimization: selecting “a subset of features that contained the maximum amount of unique information.”

The quantum effort runs through Wellcome Leap’s Quantum for Bio program along with the quantum computing company Infleqtion and colleagues at the University of Chicago and Massachusetts Institute of Technology. The team advanced to the final third phase of the program, testing a hybrid quantum-classical workflow on real quantum computers. Its current project remains exploratory, forecasting treatment response in head and neck cancer from a University of Chicago cohort. It does not establish a quantum pathway to oncology or imply that quantum systems are necessary. It also does not currently impact AI efforts to treat cancer. Its significance, like the advancements before it, is narrower. Right now, it is working toward solving an incremental problem tied directly to Pearson’s patients. Pearson is attempting to connect a distant frontier back to his own clinical work the only way he knows how: brick by brick.

Join the Conversation at SC26

SC26 brings together the people building a future that connects technology to real solutions for patients. 

Attendees engage with technical presentations, papers, workshops, tutorials, posters, and Birds of a Feather (BoF) sessions covering quantum computing, AI, and the biomedical applications of high performance computing, alongside experts from research institutions, industry, and academia working on the same problems. Pearson will also be part of the SC26 Regional Tech Hub, a multimedia forum making its conference debut in Chicago. Pearson will contribute to the Hub’s Human Impact/Translational Biomedical Research area, which will focus on how advanced computing serves society by transforming research, medicine, and healthcare.

Join us in Chicago, November 15–20, 2026, for a week of discovery, collaboration, and community with a spotlight on regional technology leaders. Register today!

Sponsored by: