Inside the Effort to Turn Tumors Into Data

Spatial biology and AI are turning archived tumour samples into rich datasets for cancer research

13 Mar 2026

Many hospitals possess a quiet treasure: vast archives of preserved tumour tissue. For decades these samples have sat in storage, used sparingly and often analysed with older laboratory techniques. A new research collaboration announced on February 4th aims to give them a second life by turning them into high-resolution datasets for artificial intelligence.

The project will apply spatial-biology technologies to archived tumour tissue. These tools measure gene activity directly within intact samples, preserving the structure of the tumour. Traditional genomic analysis usually separates cells from their surroundings, losing the context in which they operate. Spatial methods instead reveal where different cells sit and how they interact.

That detail matters. Tumours are complex ecosystems made up not only of cancer cells but also immune cells and supporting tissue. Understanding how these elements are arranged, and how they influence one another, may help researchers detect patterns that standard molecular datasets miss. Such insights could lead to new biomarkers and drug targets.

The data produced will be multimodal, combining molecular profiles with spatial information and computational analysis. The first phase of the project focuses on breast-cancer samples, with plans to expand later to lung and pancreatic tumours. Researchers hope the resulting datasets will help train predictive models useful for both biomedical research and earlier disease detection.

Yet turning archived tissue into reliable training data is not straightforward. Clinical samples collected over many years often vary in quality and preparation. Such inconsistencies complicate efforts to standardise datasets for machine learning. Researchers must also ensure that findings derived from experimental datasets hold up in real clinical settings.

Still, the effort reflects a wider shift in biomedical research. Progress in AI-driven drug discovery depends less on clever algorithms than on the quality and scale of the biological data used to train them. If projects like this succeed, the old shelves of tumour samples may become one of medicine’s most valuable data resources.