NEW YORK – Members of the international Human Cell Atlas consortium have used artificial intelligence to profile shifting cell interactions contributing to cancer based on large expression and spatial transcriptomic datasets, providing clues to breast or lung cancer development and responses to treatments.
"Having a huge amount of data about the human body is crucial to finding new ways to understand, prevent, and treat disease," Sebastian Birk, a researcher at Helmholtz Center Munich's Institute of AI for Health and the Wellcome Sanger Institute, said in a statement. "However, we also need tools that allow us to access all the benefits this information could provide."
As they reported in Nature Genetics on Tuesday, Birk and his colleagues from the Wellcome Sanger Institute, Helmholtz Center Munich, and other centers in Europe and the US developed a graph-based deep-learning tool known as "Niche identification based on cellular graph embeddings of communication programs aligned across spatial samples" (NicheCompass) to tease out and interpret cell-cell communication patterns and underlying molecular processes in simulated and real datasets.
"Unlike existing methods, NicheCompass quantitatively characterizes niches based on communication pathways," the authors wrote, noting that the tool "provides a scalable framework for identifying and analyzing niches through signaling events."
In a statement, Wellcome Sanger Institute researcher Mohammad Lotfollahi, a co-senior author on the study, likened the between-cell interactions and communication patterns that NicheCompass interrogates to the diverse types of information that individuals share with one another via social networks.
"NicheCompass is the first AI model of its kind that can interpret these networks and answer questions that could directly impact patient lives, such as highlighting where and how health conditions have started, and predicting how they might respond to certain treatments," Lotfollahi explained.
The team analyzed sequential fluorescence in situ hybridization data, spatial ATAC-RNA sequencing, and other data generated on up to 8.4 million cells in a mouse brain atlas. Moreover, they used NicheCompass to assess 10x Genomics Xenium-based spatial transcriptomics data on more than 286,500 human breast cancer cells and NanoString CosMx gene expression data for 800,559 cells from eight non-small cell lung cancer tissue samples from five donors.
"In human breast and lung cancer, NicheCompass decodes the tumor microenvironment, capturing donor-specific spatial organization and cellular processes," the authors reported, "and enables spatial reference mapping, contextualizing query datasets with a reference to identify novel niches and contrast cellular processes."
When it came to lung cancer, the team identified tumor-immune cell interaction niches that were conserved and distinct between different NSCLC patients, including an NSCLC case marked by distinct neutrophil chemoattractant features despite a comparable spatial cell arrangement.
"This real-world application not only uncovered new information that adds to our collective understanding about cancer, it also highlighted one patient whose cancer interacted with the immune system differently," co-senior author Carlos Talavera-López of the University of Würzburg said in a statement.
"In the future, NicheCompass could help uncover new ways to harness the immune system in certain cancers, creating personalized treatments that empower a patient’s immune system to target the cancer mechanisms directly," he added.