Deep within the bone marrow, blood stem cells carry the quiet record of our aging — written not in symptoms, but in the invisible architecture of their nuclei. Researchers in Barcelona have taught an artificial intelligence to read that record, developing a tool called ChromAgeNet that detects aging signatures in cellular DNA organization with 77% accuracy. The work, born from a collaboration between stem cell biologists and computational scientists, opens a new way of measuring biological time — and perhaps, one day, of slowing it.
AI tool detects aging signatures in blood stem cells from nuclear images
Aging leaves a legible signature in nuclear structure
Mark
So the AI is looking at pictures of cells and saying which ones are old. How is that different from a pathologist doing the same thing?
Mimi
A pathologist would be looking at the same microscopy images, but the changes ChromAgeNet detects are too subtle for the human eye to reliably spot. The AI found patterns in how chromatin is organized—the three-dimensional folding of DNA—that correlate with aging but aren't obvious to look at.
Luke
But 77 percent accuracy means it gets one in four wrong. Is that good enough to actually use in a lab?
Mimi
It's better than their previous computational approach, and it outperforms what you'd expect from chance. But you're right—for clinical use, you'd probably want higher confidence. For research screening, though, it could still accelerate things.
Mark
What did they actually find about aging? Like, what changed in the cells?
Mimi
Three things: chromatin entropy increased, heterochromatin moved to the edge of the nucleus, and certain chromatin condensates appeared. Those are the signatures the AI learned to recognize.
Luke
And those changes—are they known to cause the loss of function, or are they just correlated with it?
Mimi
That's the thing. The study shows the changes exist and that they're detectable. It doesn't prove they cause aging or that reversing them restores function.
Mark
They tested drugs on old cells. Did any of them work?
Mimi
The drugs changed the chromatin organization in ways that looked more like young cells. But the researchers were careful to say that doesn't mean the cells actually regained their ability to produce blood.
Luke
So it's a tool for screening, not proof of rejuvenation.
Mimi
Exactly. It's a way to quickly identify which compounds might be worth studying further. The real test would come later.
Mark
Why does this matter for actual patients?
Mimi
Blood production declines with age, which contributes to anemia and immune problems. If you could identify drugs that slow or reverse that decline, it could help older people. This tool could speed up that search.
O Pulso
- Blood stem cell aging is notoriously difficult to measure because its earliest signs are structural changes too subtle for any human eye to catch.
- ChromAgeNet, trained on thousands of 3D microscopy images, learned to distinguish young cells from old ones by reading disorder in DNA folding, heterochromatin positioning, and chromatin condensates — features no prior tool had unified into a single aging signal.
- When researchers exposed aged stem cells to epigenetic drugs and ran them through ChromAgeNet, the tool detected shifts in nuclear architecture, suggesting it could rapidly flag which therapies are worth pursuing — even before functional outcomes are known.
- The approach is built on cheap, widely available staining techniques and a lean computational model, making it scalable enough to screen thousands of samples in parallel.
- The team has released both their dataset and the model publicly, accelerating a field that has long lacked standardized tools for studying blood cell aging at scale.
Deep within the bone marrow, blood stem cells carry the quiet record of our aging — written not in symptoms, but in the invisible architecture of their nuclei. Researchers in Barcelona have taught an artificial intelligence to read that record, developing a tool called ChromAgeNet that detects aging signatures in cellular DNA organization with 77% accuracy. The work, born from a collaboration between stem cell biologists and computational scientists, opens a new way of measuring biological time — and perhaps, one day, of slowing it.
Blood production quietly falters with age. The stem cells in bone marrow that generate our red and white blood cells gradually lose their capacity, contributing to anemia, weakened immunity, and compounding vulnerability. Scientists have long understood this decline in broad strokes, but measuring its precise onset — and testing whether it can be reversed — has remained elusive. The changes are simply too fine for human observers to see.
A team led by Dr. Maria Carolina Florian at the Bellvitge Biomedical Research Institute and Dr. Paula Petrone at the Barcelona Supercomputing Center set out to change that. Their tool, ChromAgeNet, analyzes three-dimensional microscopy images of blood stem cell nuclei, teaching a convolutional neural network to recognize aging by the way chromatin — the molecular scaffolding that packages DNA — is arranged in space. Trained on thousands of images stained with a standard, inexpensive dye called DAPI, the system learned to classify cells as young or old with 77% accuracy, outperforming earlier approaches built on manually defined features.
The deeper significance lies in what the AI revealed about aging itself. By examining which visual patterns drove its predictions, the researchers identified three key biomarkers: chromatin entropy, the peripheral positioning of heterochromatin, and the presence of certain chromatin condensates. These are architectural shifts invisible to the naked eye — legible only to a machine trained to see them. Aging, it turns out, leaves a structural signature in the nucleus that can be read and quantified.
The team then used ChromAgeNet as a screening instrument, exposing aged stem cells to epigenetic drugs and checking whether the treatments nudged chromatin organization toward a younger state. The tool detected those shifts — not proof that function was restored, but a meaningful signal that certain compounds warrant deeper investigation. Because DAPI staining is cheap and ChromAgeNet is computationally lean, the approach can slot into high-throughput workflows, screening thousands of samples in parallel. The researchers have also released their dataset and model publicly, filling a long-standing gap in open resources for this field.
What ChromAgeNet ultimately offers is a new way of looking — precise, scalable, and unburdened by the limits of human perception. Whether it will lead to therapies that genuinely rejuvenate aging blood stem cells remains an open question. But it has made the invisible legible, and in doing so, brought the biology of aging one step closer to becoming a target we can actually aim at.
Blood production declines with age. The bone marrow and other tissues that manufacture our red and white cells gradually lose their capacity to keep pace with the body's needs, a deterioration that compounds over time and contributes to anemia, weakened immunity, and other age-related ailments. Scientists have long known this happens, but measuring exactly how and when blood stem cells age—and more importantly, whether interventions might slow or reverse the process—has remained difficult. The changes are often too subtle for human observers to detect.
A team led by Dr. Maria Carolina Florian at the Bellvitge Biomedical Research Institute in Barcelona and Dr. Paula Petrone at the Barcelona Supercomputing Center has developed a tool called ChromAgeNet that changes this equation. The artificial intelligence system analyzes three-dimensional microscopy images of blood stem cell nuclei and identifies patterns associated with aging by examining how chromatin—the DNA and proteins that package genetic material inside the cell nucleus—is organized in space. The work, published in Aging Cell, represents a collaboration spanning stem cell biology, computational imaging, and machine learning, with Pablo Iañez, a doctoral researcher at the Barcelona Institute for Global Health, playing a central role.
To build the model, the researchers photographed mouse blood stem cell nuclei using DAPI, a standard, inexpensive staining technique that highlights DNA. They fed thousands of these three-dimensional images into a convolutional neural network, a type of artificial intelligence designed to recognize patterns in visual data. The system learned to distinguish young cells from aged ones based purely on nuclear appearance. When tested, ChromAgeNet correctly classified cells as young or old 77 percent of the time—a performance that exceeded an earlier machine learning approach the team had built using manually defined chromatin features.
What makes this result significant is not just the accuracy but what it reveals about aging itself. The researchers analyzed which visual features the AI was actually using to make its predictions and found that the model relied on three key signatures: chromatin entropy (a measure of disorder in DNA organization), the positioning of heterochromatin at the edge of the nucleus, and the presence of certain chromatin condensates. These are not changes visible to the naked eye. They are architectural shifts in how DNA is folded and arranged—alterations so fine that only a machine trained on thousands of examples could reliably spot them. This insight matters because it suggests that aging leaves a legible signature in nuclear structure, one that can be read and quantified.
The team then tested whether ChromAgeNet could serve as a screening tool for potential rejuvenation therapies. They took aged blood stem cells and treated them with various epigenetic drugs—compounds designed to alter how genes are expressed without changing the DNA sequence itself. They then ran the treated cells through ChromAgeNet to see whether the drugs had shifted the chromatin organization toward a younger state. The results did not prove that the treatments actually restored cell function. But they demonstrated that the tool could detect changes in nuclear architecture in response to intervention, suggesting it might help researchers rapidly identify which compounds are worth pursuing further.
The practical advantages are substantial. DAPI staining is cheap and straightforward to add to existing microscopy protocols. ChromAgeNet requires relatively few computational parameters, making it fast and efficient. This combination means the tool could fit into high-throughput workflows where researchers analyze thousands of samples in parallel, dramatically accelerating the search for compounds that might preserve or restore blood stem cell function. The researchers have also released their dataset of three-dimensional cell images and the ChromAgeNet model itself to the scientific community, filling a gap in publicly available resources for studying blood cell aging.
The work points toward a future where aging is not simply observed but measured with precision, where potential therapies can be screened at scale, and where the molecular architecture of aging becomes a target for intervention. Whether ChromAgeNet will lead to treatments that actually work remains an open question. But it has created a new way of looking at an old problem—one that machines can see what humans cannot.
Citações Notáveis
The three-dimensional organization of DNA contains quantifiable information about the aging of blood stem cells and artificial intelligence can help extract it from microscopic images— Study findings, as described by research team
Quer a matéria completa?
Leia o original em Medical Xpress ↗