Chromatin strand and isolated nucleosome. DNA (yellow), histones (blue).
Credit: selvanegra / iStock / Getty Images Plus

An artificial intelligence model that analyzes how DNA is organized inside the nucleus could provide a new way to study aging in blood-forming stem cells and screen potential interventions designed to reverse age-associated changes.

Researchers from the Bellvitge Biomedical Research Institute (IDIBELL), Barcelona Supercomputing Center and collaborating institutions developed ChromAgeNet, a deep learning framework trained on three-dimensional microscopy images of hematopoietic stem cells (HSCs). The study, published in Aging Cell, suggests that changes in chromatin architecture carry enough information for AI to distinguish young from aged cells, even when those differences are difficult to identify by eye.

Reading aging from nuclear architecture

HSCs maintain blood and immune cell production throughout life, but their function deteriorates with age. Much of the search for biomarkers of biological aging has focused on molecular measurements such as DNA methylation. ChromAgeNet instead extracts information from the physical organization of chromatin, the DNA-protein complex packaged inside the nucleus.

The model was trained using DAPI-stained HSC nuclei from young and aged mice. Rather than relying on researchers to predefine which nuclear features might matter, the convolutional neural network learned patterns directly from microscopy images. Combining information across an entire 3D image stack improved its ability to separate the two age groups.

The team also investigated what was driving the predictions. Explainable AI analyses pointed to features including chromatin entropy, peripheral heterochromatin and chromatin condensates. In younger cells, informative regions tended to cluster around structures such as the nuclear envelope, whereas attention was more dispersed in aged nuclei, consistent with increasing disorganization of chromatin during aging.

A possible screening tool for rejuvenation research

The researchers next applied ChromAgeNet to aged HSCs exposed to compounds affecting Rho GTPase signaling or chromatin regulation. Several treatments shifted the cells’ AI-derived “youthful scores” toward those seen in young HSCs, with particularly clear changes following treatment with chromatin-modifying compounds.

That finding makes image-based phenotypic screening an intriguing application. DAPI staining is inexpensive and already widely used, potentially allowing the approach to fit into relatively scalable microscopy workflows rather than requiring a molecular profiling assay for every cell.

But a younger-looking nucleus is not synonymous with a rejuvenated stem cell. As the authors stress, “direct functional validation and independent measurements of biological age will be required” before changes in ChromAgeNet scores can be interpreted as genuine rejuvenation.

Translation will require broader validation

For now, ChromAgeNet remains a preclinical proof of concept. The model was developed in mouse HSCs, and chronological age was used as a proxy for biological age despite substantial cell-to-cell heterogeneity. Imaging conditions can also introduce batch effects, making independent validation across laboratories, microscopes, and experimental settings important.

Human translation presents a further hurdle because mouse and human cells differ in both chromatin organization and imaging characteristics. The researchers therefore expect that application to human HSCs will require model adaptation and independent validation.

The broader opportunity is less about creating another chronological aging clock and more about establishing an accessible phenotype that can be measured repeatedly during drug discovery. If the imaging signature can ultimately be connected to functional stem cell health, ChromAgeNet could provide researchers with a scalable way to ask not simply how old a cell looks, but whether an intervention meaningfully changes its aging state.

Also of Interest