Cancer is not just a collection of rogue cells; it is a neighborhood where cells talk to each other to build defenses. Standard tests often look at cells in isolation, missing the hidden connections that let tumors survive. This review explores a new way to see those connections using spatial multi-omics and deep learning, specifically graph neural networks. These tools act like a super-powered map, showing how cells are arranged and how they interact to resist treatment.
The study looked at how these advanced computer models could identify specific patterns, or spatial phenotypes, that show why an immune system might fail to attack a tumor. It also aimed to find topological biomarkers, which are unique signatures of the tumor's shape and structure, and understand the exact mechanisms of immune evasion. The goal is to create better tools for validating these biomarkers in the future.
But there are serious hurdles before this becomes a routine part of care. The review highlights major challenges in standardizing data so different labs can compare results. It also points out issues with computational scalability, meaning the technology might be too heavy for everyday use. Furthermore, the models often lack explainability, making it hard for doctors to trust why the AI made a specific call. Until these problems are solved and regulatory approval is granted, this remains a promising but unproven path.