AI-powered lung cancer atlas identifies cell types linked to higher risk of spread

by Ludwig Maximilian University of Munich

edited by Lisa Lock, reviewed by Robert Egan

Lisa Lock

Scientific Editor

Meet our editorial team
Behind our editorial process

Robert Egan

Senior Editor

Meet our editorial team
Behind our editorial process Editors' notes

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

fact-checked

peer-reviewed publication

trusted source

proofread

The GIST Add to Preferred Sources

Recent advances in cancer biology have shown that a tumor is not a uniform mass. It consists of many different cancer cells and is embedded in an environment of immune cells, connective tissue and blood vessels. This variety is a significant factor in determining whether a tumor spreads and how well a therapy works. While single-cell approaches are considered pioneering instruments in cancer research, their translation into clinical practice has been limited to date, primarily due to a lack of large patient data sets with comprehensive clinical histories.

Researchers from LMU Munich, Charité University Hospital Berlin and the Berlin Institute for the Foundations of Learning and Data (BIFOLD) at the Technical University of Berlin have now created a comprehensive single-cell atlas for lung cancer. This atlas was then linked to the clinical outcomes of patients using AI.

With their model, the researchers were able not only to identify tumor cell types that particularly contribute to the development of metastases but also to predict interactions between tumor cells and the cells of the surrounding microenvironment that are closely connected to disease outcome. The study has now been published in the journal Nature Cancer.

Capturing tumor cells and their environment in detail

One of the world's leading causes of cancer-related deaths, lung cancer is markedly heterogeneous. Even within a tumor, cancer cells differ considerably in their aggressiveness. The conventional classification according to tumor size and lymph node involvement does not adequately capture this variety. And although single-cell technologies can make this variety visible, there has been a lack of large data sets that connect single-cell profiles with clinical follow-up data.

"For the LuCaS+ atlas, our team analyzed tumor tissue from 454 lung cancer patients," explains professor Frederick Klauschen from LMU. "To do this, we used a method that had rarely been used before on this scale: With the help of single-cell RNA sequencing, we were able to read out the activity of over 2,000 genes per cell nucleus from archived tissue samples and thus draw on clinical histories with long follow-ups."

In total, the researchers captured the gene activity in more than 2.5 million individual cells. Using machine learning methods, they were able to find patterns within the data. This allowed them to define numerous groups of tumor cells, as well as groups of immune system, connective tissue and blood vessel cells, that correlate with clinical outcomes.

Recognizing aggressive tumor cells before metastases occur

Using these clinical data, the researchers determined the metastatic potential for each tumor cell group. This approach allows the identification of patients who have a high risk of developing metastases despite their tumors being in early stages where adjuvant therapy is not typically administered after surgery. These findings were validated in an independent patient cohort.

"When working with millions of cells, classical evaluation methods come up against their limits," says professor Klaus-Robert Müller, BIFOLD director and professor at TU Berlin. "Our machine learning methods, by contrast, recognize even the tiniest differences between individual cell types. This makes it possible to link these patterns directly to the clinical outcomes of the respective patients."

Interplay of tumor and environment opens new therapeutic avenues

In a second step, the team incorporated the microenvironment—that is, the immune, connective tissue and blood vessel cells that surround and interact with the tumor. From the joint analysis of the metastatic potential of tumor cells and their specific microenvironment, the researchers developed a model that predicts clinical outcomes significantly better than traditional parameters.

The model recognized patients with a higher risk of relapse than conventional evaluations would indicate—and also patients whose prognosis is more favorable than previously reckoned. At the same time, it showed which interactions between tumor and microenvironment influence the course of the disease.

"LuCaS+ is freely available to scientists. The goal is to derive new therapeutic approaches and treatment strategies, which will then have to be tested in cell culture and animal models," says Dr. Simon Schallenberg of Charité.

Publication details

Philipp Jurmeister et al., A single-cell lung cancer atlas reveals cellular and molecular drivers of metastasis and clinical outcome, Nature Cancer (2026). DOI: 10.1038/s43018-026-01260-w

Journal information: Nature Cancer

Key medical concepts

Lung CancerNeoplasm MetastasisTumor Microenvironment

Clinical categories

OncologyPulmonary medicine Provided by Ludwig Maximilian University of Munich Who's behind this story?

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021. Full profile →

Robert Egan

Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →

Citation: AI-powered lung cancer atlas identifies cell types linked to higher risk of spread (2026, November 12) retrieved 13 November 2026 from https://medicalxpress.com/news/2026-11-ai-powered-lung-cancer-atlas.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.