Researchers Develop AI Tool that Maps Cellular "Social Networks" to Guide Cancer Treatment
Researchers have developed a new artificial intelligence (AI) tool that can analyze how cells communicate within tissues, potentially guiding more personalized treatments for cancer and other diseases.
The tool, called NicheCompass, was created by scientists from the Wellcome Sanger Institute, Helmholtz Munich's Institute of AI for Health, the University of Würzburg, and collaborators as part of the Human Cell Atlas Initiative. According to the study, published in Nature Genetics, NicheCompass can quickly interpret complex cellular networks, helping researchers understand how diseases develop and how different patients might respond to treatment.
"People often communicate to their networks with a range of different information. They might share developments from work, or pictures of their holidays, and while these might be to different friends, they can all be traced back to one individual. Cell-to-cell communication is similar, cells might use different features to communicate with their social network, creating communities or networks in their local area."
— Dr Mohammad Lotfollahi, co-senior author at the WSI.
NicheCompass is designed to analyze the "social networks" of cells—the way cells interact with each other based on shared surface proteins and molecular signals—by modeling how these interactions shape functional communities, or niches, within tissues. In biological terms, niches are small, diverse groups of colocalized cells that coordinate specific functions within tissues. These cellular communities are defined by shared signaling events and gene expression patterns, which reflect how cells communicate and adapt to their environment.
See also: 'Google Maps' of Human Cells
The AI model combines single-cell and spatial genomic data to create a detailed map of cellular neighborhoods, allowing it to visualize where different cell types are located and how they interact at a molecular level. While existing computational methods tend to cluster cells based on gene expression alone, the idea behind NicheCompass is to use deep learning to align similar networks of cells and identify signaling-based patterns. This allows it to characterize niches not just by their location but also by the specific signaling pathways that drive cellular behavior. The model learns from known molecular signaling events but can also discover new patterns of co-expressed genes and chromatin features, expanding its ability to identify and characterize novel niches.
An overview of the tool: It analyzes spatial omics data to map how cells interact with each other within tissues. It creates a graph where each cell is a node, capturing both the cell's location and molecular features like gene expression. A graph neural network (GNN) processes this data to identify patterns in how cells communicate, adjusting for differences between samples. The model learns both known and new interaction pathways, helping researchers understand how genes and nearby cells influence each other. It can then reconstruct these spatial and molecular relationships, giving insights into how cells behave in healthy and diseased tissue.
In the study, researchers used NicheCompass to analyze data from 10 lung cancer patients. The tool identified both similarities and differences in how cancer cells communicated with their surroundings, offering clues for potential treatment targets. It also highlighted how one patient's immune cells interacted differently with the tumor, suggesting possible strategies for leveraging the immune system in treatment. The team also applied NicheCompass to breast cancer tissue and demonstrated its ability to detect complex cellular interactions across different cancer types. According to the authors, the model can process patient data and generate meaningful findings within one hour.
Beyond cancer, NicheCompass was tested on a mouse brain spatial atlas containing 8.4 million cells. The model correctly identified brain sections and created a visual resource of the entire organ, showing its potential to analyze complex tissue structures in other areas of biology. Dr. Carlos Talavera-López, a co-senior author from the University of Würzburg, said that this capability could lead to more personalised targeted treatments.
"Having a huge amount of data about the human body is crucial to finding new ways to understand, prevent, and treat disease. However, we also need tools that allow us to access all the benefits this information could provide", said Sebastian Birk, first author from Helmholtz Munich and the Wellcome Sanger Institute.
Topics: AI & Digital
