Stanford and Mayo Clinic Use AI Blood Tests to Map Tumor “Neighborhoods”

Stanford Medicine and Mayo Clinic researchers have developed an artificial intelligence framework that uses blood-based liquid biopsies to map the cellular environments surrounding tumors, a step that could expand precision oncology beyond tumor DNA analysis alone. The research was published in Nature earlier this month.

The system combines two machine learning models called Spatial EcoTyper and Liquid EcoTyper. Together, they analyze methylation signals in cell-free DNA from blood samples to reconstruct tumor microenvironments, including immune and stromal cell activity linked to therapy response.

Current FDA-cleared liquid biopsy workflows largely focus on circulating tumor DNA mutations from cancer cells themselves. The Stanford-Mayo approach instead attempts to profile the broader tumor microenvironment, which researchers say plays a major role in determining whether patients respond to immunotherapy.

The researchers identified nine recurring “spatial ecotypes,” or cellular neighborhoods, across 17 solid tumor types, including melanoma, lung, gastric, bladder, and breast cancers. Some ecotypes were associated with favorable immunotherapy responses, while others correlated with treatment resistance and poorer survival outcomes.

Aaron Newman, Associate Professor of Biomedical Data Science at Stanford Medicine, said the models were intentionally designed to avoid black-box clinical predictions by exposing which methylation markers influenced the system’s outputs.

The focus on explainable clinical AI comes as hospitals and healthcare systems place greater emphasis on governance, transparency, and physician oversight for deployed models, including broader efforts around explainable AI in healthcare operations.

Blood-Based Monitoring Expands Precision Oncology Research

The researchers said the system could eventually allow oncologists to monitor how tumors evolve during treatment using serial blood draws rather than repeated invasive tissue biopsies.

In early patient monitoring studies, the researchers observed shifts in tumor microenvironment signals months before changes appeared on imaging scans.

Dr. Aadel Chaudhuri, a radiation oncologist at Mayo Clinic and professor in the Department of Radiation Oncology, said the framework attempts to address a major limitation in current oncology biomarker systems.

“One of the challenges is that these existing methods only have modest associations,” Chaudhuri said in Mayo Clinic’s announcement. “They’re essentially ‘surrogates of surrogates’ and don't fully capture what's happening inside the tumor environment.”

The researchers also said the system outperformed conventional biomarkers such as PD-L1 expression and tumor mutational burden in several evaluated patient cohorts.

The work arrives as healthcare organizations continue expanding AI deployments across diagnostics, imaging, genomics, and hospital operations. Recent healthcare AI initiatives have included systems for cardiovascular risk detection from CT scans and broader operational AI deployments across hospital networks.

The Stanford-Mayo research remains in the clinical study phase. Researchers said they are continuing to evaluate how tumor microenvironments evolve under immunotherapy, CAR-T therapy, antibody-drug conjugates, and bispecific therapies.

The team said its long-term goal is to make Liquid EcoTyper-based analyses broadly available by 2030.