AI reads telomere length from biopsy slides
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Sanju Sinha, Ph.D., and colleagues at Sanford Burnham Prebys Medical Discovery Institute developed TLPath, a model that infers telomere length from structural features in routine histopathology slide images, according to findings published in Cell Reports Methods.
- TLPath was trained on 5,263 histopathology slides spanning 18 tissue types from 919 donors, drawn from the NIH's Genotype-Tissue Expression (GTEx) Project.
- The model segments each slide into an average of 1,387 patches and extracts up to 1,024 structural features per patch, building on recent advances in histopathology foundation models for computer vision.
- TLPath outperformed age-based telomere predictions and successfully distinguished telomere length differences between individuals of identical chronological age.
- Sinha identified the limiting factor as data access, not modeling: standard clinical biopsy slides are rarely digitized and shared with researchers in the way GTEx data is, despite being routinely produced in patient care.
- The paper, titled "Tissue Morphology Predicts Telomere Shortening in Human Tissues," appears in Cell Reports Methods (2026) with DOI 10.1016/j.crmeth.2026.101336.
Why it matters: Direct telomere length measurement requires complex, costly lab tests that don't scale to large populations, despite telomere length being linked to chronic disease risk and aging. TLPath repurposes slides already generated in routine clinical care, potentially turning the existing biopsy pipeline into a scalable aging-research tool—if hospitals and biobanks start scanning and sharing those slides.
