UAB O’Neal researchers publish AI approach to mapping genetic differences within tumors
Researchers at the University of Alabama at Birmingham (UAB) and the UAB O’Neal Cancer Center are part of a new study published in Nature Biomedical Engineering that uses artificial intelligence to uncover genetic differences within tumors from images routinely produced during cancer diagnosis.
Published August 24, 2026, the title of the study is “HisToSpatialCNV: an interpretable deep learning method predicting spatial copy number variations from histopathology images.”
Lana Garmire, Ph.D., FAIMBE, FACMI, professor and vice chair of research in the Department of Biomedical Informatics and Data Science at the UAB Heersink School of Medicine and senior scientist at the O’Neal Cancer Center, is the corresponding author. Other UAB researchers listed as authors include Xuhui Guo, Ph.D., and Pooja Thakur, Ph.D.
The research introduces HisToSpatialCNV, a deep-learning method that analyzes H&E-stained pathology images to predict spatial copy number variations (CNVs).
Finding genetic differences within a tumor
A pathology slide can reveal a great deal about how cancer cells look and how they are organized within tissue. What it cannot show directly are many of the genetic changes taking place inside those cells.
Garmire and her team developed HisToSpatialCNV to help researchers investigate those changes using information contained in standard pathology images.
“HisToSpatialCNV was developed to help bridge that gap,” Garmire said. “The model combines interpretable image analysis with graph neural networks and attention-based approaches to analyze both local and broader tissue characteristics.”
UAB researchers evaluated the approach using breast, skin, and brain cancer datasets. They found that the CNV patterns predicted from the pathology images showed strong agreement with spatial copy number variation information, moreover the spatial copy number variations proceed the pathology detectable tumor vs non-tumor differences.
Additionally, the model provided a way to examine genetically distinct groups of cancer cells within the same tumor clinical subtype.
“In a HER2-positive breast cancer dataset, for example, we identified four distinct tumor subclones with different patterns of genetic alterations, pathway activities and patient outcomes,” Garmire said.
A new way to study tumor biology
Tumors are not made up of identical cancer cells. Different regions can contain cells with different genetic alterations and biological behaviors, a phenomenon known as tumor heterogeneity.
Read the full article about Garmire’s study online at https://www.nature.com/articles/s41551-026-01754-z