AI reads aging in blood stem cell nuclei
You can tell how biologically old someone is by looking inside a blood stem cell. Researchers trained an algorithm on microscopy images of cell nuclei and used it to build a new aging clock.
Blood stem cells, known as hematopoietic stem cells (HSCs), are the cells from which all blood cells originate. They reside in the bone marrow and remain active throughout life. As they age, their interiors change: the way their DNA is packaged and arranged inside the cell nucleus shifts over time. These changes in chromatin structure, the combination of DNA and proteins that organizes genetic material, are visible under a microscope but far too complex to analyze manually.
The researchers developed a system called ChromAgeNet, based on convolutional neural networks. They trained the algorithm on three-dimensional microscopy images of cell nuclei from young and old mice. The system learned to distinguish between young and aged stem cells, reaching an accuracy score (AUROC) of 0.77, which the researchers consider a solid proof-of-concept result.
What does the AI actually see?
The algorithm identified three features as predictors of biological age: the degree of disorder in chromatin organization (chromatin entropy), the distribution of densely packed DNA near the nuclear membrane (peripheral heterochromatin), and the presence of condensed chromatin clusters. All three shift as a stem cell ages.
A secondary finding was that the model could detect whether aged stem cells had been exposed to epigenetic drugs. This makes the system potentially useful as a screening tool: researchers could use it to test candidate compounds for their ability to rejuvenate stem cells, without the need for expensive and time-consuming functional assays.
A clock that screens for rejuvenation
Aging clocks come in many forms: based on DNA modifications, protein profiles, or image analysis. This image-based approach adds a new dimension to the field. Because the findings are from mouse models, caution is warranted when extrapolating to humans. But the approach reveals a principle: the spatial organization of DNA inside a cell nucleus tells a story about biological age that algorithms can be trained to read.
Want to research this yourself?
Search for example:
- hematopoietic stem cell aging
- chromatin structure aging clock
- deep learning biological age