Tumor floor plans may predict immunotherapy response in lung cancer

Inside a tumor biopsy, the decisive clue may be less a single gene than the way thousands of cells arrange themselves. At the University of Chicago, Arjun Raman’s team has built statistical “floor plans” of tumors from spatial transcriptomics and used them to predict which non-small cell lung cancer patients would respond better to immunotherapy than the current standard-of-care biomarker.

Spatial transcriptomics records both gene activity and its location inside tissue. That matters because tumor cells do not act alone: they gather into local neighborhoods, or spatial groups, and those groups interact to form larger structures. Raman compares the pattern to a flock of birds—individual behavior contributing to a collective shape.

The team applied this framework to published data from 262 solid tumors. Instead of comparing tumors only by their cell populations or genes, the researchers compared their spatial groups and placed them in a statistical “latent space,” an artificial-intelligence representation in which more similar tumor layouts sit closer together.

The test involved 16 tumor samples from non-small cell lung cancer patients, supplied by Marina Garassino’s laboratory at UChicago, plus more than 200 other tumors and clinical information about treatment response. The researchers reported that the layout-based comparison correctly predicted immunotherapy response better than the existing standard biomarker. The measurements come from the research team’s analysis; the source does not report independent replication or a clinical trial.

So what changes, concretely? If the approach holds up in further studies, an oncologist could eventually profile an individual biopsy, compare its organization with known tumor patterns and obtain a treatment-response estimate within hours. That remains a future goal, not a service currently available: Raman is extending the work to ovarian cancer and chemotherapy, while also exploring synthetic reference patterns that other researchers could use for comparison.