Lung cancer study links AI-mapped cell groups to higher risk of spread

A new lung cancer atlas connects gene activity inside individual cells with patients’ long-term outcomes. Published in Nature Cancer on November 12, 2026, the LuCaS+ study identifies tumor-cell groups associated with cancer spread. Researchers also report better survival prediction when they analyze cancer cells alongside surrounding tissue, with findings validated in an independent patient cohort.

LuCaS+ contains profiles from more than 2.5 million cell nuclei drawn from 454 people with non-small-cell lung cancer. The researchers paired those measurements with detailed clinical records and long-term follow-up. That connection addresses a central obstacle in single-cell cancer research: making detailed biological maps useful for understanding what happens to patients.

The team used single-nucleus RNA sequencing, which measures gene activity in a cell’s nucleus. Working with preserved, archived tissue let them study samples whose associated patient histories were already available. According to Frederick Klauschen of LMU Munich, the measurements captured activity from more than 2,000 genes per nucleus.

Machine learning helped organize the cellular variation. The researchers used contrastive learning and hierarchical clustering to identify 500 cell clusters, including 136 classified as malignant. They caution that each malignant cluster does not necessarily represent a distinct cancer-cell subtype; the fine-grained grouping preserves subtle differences for further analysis.

364 patientsDiscovery cohort

Samples from Berlin and Cologne formed the dataset used to develop the findings.

90 patientsIndependent validation cohort

Samples from Munich were processed separately and held out for validation.

The study identified tumor-cell populations with high metastatic potential, meaning a greater association with cancer spreading. Their molecular features included cancer-promoting signaling, evasion of the immune system and changes associated with cells becoming more mobile. These populations had strong prognostic value in early-stage lung adenocarcinoma, one of the cancer types studied.

The researchers describe a possible use: identifying patients at high risk of metastasis even when their tumors are at an early stage. LMU’s account notes that additional treatment after surgery is not typically given in the early-stage setting it describes. The cell patterns could therefore expose risk that conventional assessments miss.

Cancer cells were only part of the picture. The atlas also distinguished 60 nonmalignant cell types, including immune cells, connective-tissue cells and cells associated with blood vessels. Together, these form the tumor microenvironment: the surrounding tissue that interacts with the cancer rather than merely sitting beside it.

Combining tumor and surrounding-cell profiles improved survival prediction across clinical stages, the authors report, adding prognostic information beyond current clinical standards. The analysis also distinguished different biological settings: inflammation involving monocytes and fibroblasts in adenocarcinoma, blood-vessel-growth signaling in squamous cell carcinoma, and environments favorable to antitumor immune activity.

The atlas is available for scientists to explore, including through lucas-plus.com. Its interactive resource brings together cell annotations, gene expression, cellular composition and clinical information. The paper also provides analysis code through GitHub and makes sequencing data and cell-level annotations available through Zenodo for noncommercial use.

That access gives other researchers a way to investigate the links between cellular biology and clinical outcomes, rather than relying only on the paper’s selected findings. The authors present LuCaS+ as a resource for investigating disease mechanisms and potential drug targets, with patient follow-up anchoring those questions in observed outcomes.

The treatment implications remain hypotheses. In LMU’s account, Charité researcher Simon Schallenberg says the goal is to derive new therapeutic approaches and treatment strategies, which must then be tested in cell cultures and animal models. The independent patient validation supports the outcome findings; it is not a demonstration that acting on them improves treatment.