Researchers have developed an artificial intelligence tool called ChromAgeNet that reads microscopic 3D images of mouse blood stem cells to distinguish younger DNA patterns from aged ones. Published in Aging Cell, the technology offers a novel way to assess cellular aging and screen potential rejuvenation compounds.
As living organisms grow older, the hematopoietic system—the network of organs and tissues responsible for blood production—gradually becomes less able to produce an adequate supply of blood cells. Preserving or restoring this blood production requires a deep understanding of how hematopoietic stem cells age, yet the physical transformations within these microscopic nuclei are often too subtle to recognize under a microscope. To overcome this limitation, a research team built an AI system designed to read the hidden architectural signatures of cellular time in mouse blood stem cells.
How ChromAgeNet Reads Nuclear Architecture in Blood Stem Cells
Inside every cell nucleus, genetic material is packaged into chromatin, a material made mainly of DNA and proteins that helps regulate which genes are active and shapes overall cellular identity. Rather than relying on traditional chemical markers like methylation, the new model evaluates how DNA folds and organizes in three dimensions inside individual cells. The new system outperformed an alternative machine learning approach that depended on predetermined chromatin features established by the team.
The collaborative project was led by Dr. Maria Carolina Florian, an ICREA Research Professor in the Regenerative Medicine Program at the Bellvitge Biomedical Research Institute (IDIBELL), alongside Dr. Paula Petrone from the Barcelona Supercomputing Center–Centro Nacional de Supercomputación (BSC-CNS) and the Barcelona Institute for Global Health (ISGlobal). Their team gathered three-dimensional microscope images of mouse hematopoietic stem cell nuclei stained with DAPI, a routine and inexpensive technique used to make DNA visible. A convolutional neural network then learned to separate young cells from aged ones, achieving a 77% probability of correctly classifying them in a study published in Aging Cell.
Petrone noted that the artificial intelligence can spot subtle age-related differences that are not necessarily perceived by the human eye.
Unlike opaque algorithms that obscure their reasoning, the neural network allowed investigators to inspect which physical characteristics drove its predictions. Among the most influential clues were chromatin entropy, which measures disorder, and heterochromatin concentrations located near the edge of the nucleus, as well as distinct chromatin condensates. Examining which features of the images helped the model make its predictions allowed scientists to understand what physically changes inside cells as they age, rather than simply getting an age prediction from a black-box algorithm.
Nuclear Architecture Shifts Do Not Prove Cellular Rejuvenation
Beyond simply estimating biological age, the team explored whether the model could track cellular responses to pharmacological interventions.
While certain treated cells developed structural patterns resembling younger controls, the investigators emphasized an important caveat. Shifting nuclear architecture toward a younger configuration does not prove rejuvenation or establish that the cells recovered their lost biological function.
Because DAPI staining is inexpensive and straightforward to integrate into microscopy protocols, the compact design of the model could facilitate high-throughput microscopy of large quantities of samples. This capability may eventually help laboratories identify drugs capable of improving the immune system and the quality of life of older people.
The underlying research formed a central part of the doctoral thesis of ISGlobal researcher Pablo Iañez, combining expertise across stem cell biology, aging, image analysis, and artificial intelligence. To foster further investigation, the team has released ChromAgeNet to the scientific community alongside a dataset of three-dimensional hematopoietic stem cell images.