AI recognizes signs of stem cell aging

The development was presented by researchers from the Bellvitge Institute for Biomedical Research (IDIBELL) and the Barcelona Supercomputing Center. ChromAgeNet is an artificial intelligence‑based model that, based on microscopic images, determines whether the hematopoietic stem cell in front of us is young or old. The secret lies in the fact that it analyzes the three‑dimensional organization of chromatin — the complex of DNA and proteins inside the nucleus. It is the structure of chromatin that largely determines which genes will be active, and therefore also determines the cell’s behavior.

To train the model, the researchers used three‑dimensional images of mouse stem cell nuclei stained with the inexpensive and widely available DAPI dye — it highlights DNA. The neural network learned to distinguish age‑related features from these images. ChromAgeNet is based on a convolutional neural network — a type of AI specifically designed for image analysis. As a result, the model learned to determine the cell’s age with a probability of 77 %, surpassing the accuracy of the previous approach, in which aging features were defined manually.

The most important thing here is that age‑related changes in the nucleus are often too subtle for a person to reliably detect. Stem cell aging is an uneven process, and the reorganizations of the nuclear architecture can be barely perceptible. ChromAgeNet, however, is able to identify combinations of spatial features of chromatin that together carry information about the cell’s age. The researchers further analyzed which specific details the model relies on most often: these are chromatin entropy, the location of heterochromatin at the edge of the nucleus, and individual dense chromatin clumps. Knowledge of these markers not only helps to distinguish between young and old cells — it reveals the very picture of how the architecture of the nucleus changes with age.

This approach could be a good addition to existing methods for assessing biological age, for example, to “epigenetic clocks,” which determine age based on chemical changes in DNA. But ChromAgeNet has another use: it can be used as a tool to find ways to rejuvenate cells. As a pilot experiment, scientists tested how the model responds to treatment of old stem cells with various epigenetic drugs. The goal was not to prove real rejuvenation, but to see whether the signs of aging in the chromatin structure change in such a way that it appears as if the cell has become younger. The results showed that ChromAgeNet is indeed capable of detecting such shifts, which means it can help in identifying promising strategies.

Another advantage of the method is its compatibility with high‑throughput workflows. DAPI is a cheap and easy‑to‑use dye, and the model itself contains relatively few parameters. It is convenient to integrate it into protocols where it is necessary to quickly analyze thousands of samples. In the long term, this will speed up the testing of various substances and help to more quickly identify candidates capable of preserving or restoring stem cell functions.

Furthermore, the authors of the study have made publicly available a set of three‑dimensional images of hematopoietic stem cells and the ChromAgeNet model itself. This is a valuable resource for the scientific community; there are almost no such open image sets, and they are needed to develop and test new computational methods for studying cell aging.

Overall, the work shows that the three‑dimensional organization of DNA contains measurable signs of aging in hematopoietic stem cells, and artificial intelligence helps to extract these signs from ordinary microphotographs. ChromAgeNet opens up the possibility of studying cellular aging faster, on a larger scale, and more accurately, and in the future it may become an important link in the search for ways to maintain the health of the hematopoietic system as it ages.