A new computational tool infers changes occurring at the ends of the chromosomes housing our DNA. It does so by detecting structural alterations in cells and tissues captured in images taken of routine medical biopsies, according to findings published March 16, 2026, in Cell Reports Methods.
Scientists at Sanford Burnham Prebys Medical Discovery Institute developed the TLPath model based on the hypothesis that modifications in the shape and structure of cells and tissues could be used to predict the length of repeating sections of DNA called telomeres.
"Whenever DNA gets replicated as our cells grow and divide, the part at the end of the DNA cannot be replicated," said Sanju Sinha, PhD , an assistant professor in the Cancer Metabolism and Microenvironment Program at Sanford Burnham Prebys.
"This would be a problem if our DNA was degraded bit by bit from birth, but instead our cells evolved a unique solution of capping the ends of DNA with repeating regions called telomeres that can be whittled down instead of more essential genetic information."
Telomeres, however, are not mere genetic buffers to be freely discarded. While scientists are still determining exactly how these DNA bumpers affect the aging process, researchers have found that the length of telomeres is correlated with a person's chronological age throughout their lifespan. After tracking health outcomes in large populations, telomere length was found to predict patients' risk of chronic diseases associated with aging.
"We were reasonably certain that telomeres play an important role as cells age, and we knew the field needed more ways to study this phenomenon to learn how it can be treated to benefit patients," said Sinha.
The research team obtained data from the Genotype-Tissue Expression Project, a major National Institutes of Health (NIH) Common Fund initiative that launched in 2010 to create a resource for studying how inherited changes in genes lead to common diseases. Sinha and his colleagues were able to train their computational model on scans of 5,263 histopathology slides made from routine biopsy samples of 18 tissue types that were donated by 919 individuals.
"The dataset pairs these high-resolution images with laboratory tests of telomere length, enabling us to train TLPath to find predictive features in the cells and tissue," said Sinha. "There are hundreds of terabytes of imaging data from this project ripe for study with tools such as TLPath, and we could not have finished our project without this data being available to researchers."
The model works by segmenting each histopathology slide into an average of 1,387 square fragments. Each fragment, known as a patch, is scoured to find up to 1,024 structural features. By computing a statistical weight for each feature on each patch, the model compares an overall score for each histopathology slide with the paired telomere length to learn how to predict the latter from the former.
After training TLPath separately on each tissue type, the scientists found it capable of predicting telomere length on samples from the Genotype-Tissue Expression Project that had not been included in the training dataset.
"The key to our work was building on recent developments in computer vision for histopathology slides, which is the creation of foundation models," said Sinha. "These models don't look at discrete pixels, but instead define more higher order features, only some which can be interpreted by humans yet can be validated for their predictive power."
In testing, TLPath succeeded in more accurately predicting telomere length than basing the prediction solely on the age of patients when they donated their samples. The scientists further evaluated the model's prediction capabilities by demonstrating that it could identify telomere length differences between individuals of the exact same chronological age.
"This opens up new opportunities based on the conceptual advancement that measurable structural changes in cells can predict the length of telomeres," said Sinha. "Directly measuring telomere length requires more complicated and costly tests that are difficult to scale.
"The only limit to using an approach such as TLPath is the availability of scanned histopathology slides."
While these slides are commonly created from biopsies for pathologists to review in the course of clinical care, they are rarely digitized and made available to researchers in a similar manner as the NIH-funded Genotype-Tissue Expression Project.
"Whether it is new slides being developed today or those preserved in biobanks, all we need is for them to be properly scanned, stored and shared in order to enable large-scale studies," said Sinha.
"This has the potential to transform our ability to study telomere biology, learn more about human aging and ultimately help people preserve better health as they age."
Anamika Yadav, a research assistant at Sanford Burnham Prebys, shares first authorship of the study with Kyle Alvarez, a graduate student at Sanford Burnham Prebys.
Additional authors include Akanimoh Adeleye, Yu Xin Wang and Michael Jackson from Sanford Burnham Prebys.
The study was supported by the National Cancer Institute-designated Cancer Center at Sanford Burnham Prebys and the Lab Experience As Pathway to Graduate School Program at Sanford Burnham Prebys.
The study's DOI is 10.1016/j.crmeth.2026.101336.
