AI study links tumor signals and protein barriers to immune-cell engagement

A new AI framework connects the arrangement of cells in tumors with molecular signals associated with immune engagement. Researchers introduced SPACER in Nature Biomedical Engineering on October 13, 2026. The study links local gene activity and potential protein barriers to T-cell engagement, offering clues to why some tumors respond poorly to immunotherapy.

From cell locations to recruitment signals

Researchers at the University of Texas MD Anderson Cancer Center and collaborating institutions developed SPACER, short for Spatial Analysis of Cellular Engagement and Recruitment. It analyzes spatial transcriptomics: measurements that show which genes are active and where that activity occurs within intact tissue.

The distinction is between describing a tissue map and investigating the signals behind it. The paper’s authors say existing computational tools describe where cells reside but cannot explain why. SPACER is designed to infer gene-level rules associated with cell recruitment directly from those spatial measurements.

Its approach also differs from treating each cell as an isolated unit. SPACER examines groups of neighboring cells and how they work together. MD Anderson’s account describes that tissue-based organization as the reason its predictions are easier to interpret, rather than emerging from an unexplained black box.

The authors applied the framework to 37 datasets spanning several human tumor types and mouse heart tissue. They report four- to 21-fold improvements over existing methods on tasks that recover biologically meaningful recruitment signals. That comparison concerns signal discovery; it is not a reported improvement in patient treatment outcomes.

Signals that invite engagement—and barriers that resist it

Across the cancer types studied, local antigen presentation was the strongest correlate of T-cell engagement. Antigen presentation is the process through which cells display protein fragments for immune cells to inspect. Tumors expressing genes involved in that process were more likely to attract T cells.

The researchers interpret this as evidence that immune cells respond primarily to signals in their immediate surroundings, rather than signals circulating elsewhere in the body. The finding puts the tumor’s local interactions with immune cells at the center of the recruitment question.

A second finding points in the opposite direction. Mucins—large, sugar-coated proteins produced by many cancers—were associated with physical barriers to T-cell engagement. The researchers suggest these proteins may help keep T cells away from tumors, even as other local signals are associated with attracting them.

That association also tracked treatment response. Across melanoma and gastric cancer datasets, higher tumor mucin levels were associated with poorer responses to immune checkpoint inhibitors and poorer patient outcomes. These are associations with response and prognosis, not a demonstration that reducing mucins would improve treatment.

SPACER is an exciting step toward decoding that language, allowing us to connect the patterns we see in tissues to the biology that may explain them. This helps us turn these insights into new ideas we can test in the lab.

Linghua Wang, study co-leader, in MD Anderson’s account

A heart-inflammation test beyond tumors

The same study examined preclinical models of immunotherapy-related myocarditis, or heart inflammation. SPACER identified heart regions and molecular signals associated with T-cell recruitment. The analysis suggested that CD4-positive T cells may play a larger role in severe inflammation than previously recognized.

The authors have made SPACER’s code available through GitHub and provide a documentation site. The paper also identifies public repositories for study data, including the mouse heart measurements and newly generated human spatial data. Source data accompany the publication, giving researchers access to both the framework and material behind its findings.