As people age, some cells in the body enter a state of senescence: they stop dividing but do not die and can accumulate in tissues. These cells are called “zombie cells” because their accumulation may be associated with cancer, tissue degeneration, and inflammation.
A team at the Massachusetts Institute of Technology in the United States has developed an artificial intelligence method for identifying such cells without destroying them. The researchers combined biochemical data obtained with Raman microscopy with gene expression data from individual cells and used machine learning to isolate markers characteristic of senescence.
Markers used to detect senescent cells include the p16 and p21 proteins, which are involved in stopping cell division. However, detecting them usually requires damaging or destroying the cell, whereas Raman microscopy obtains information about the chemical composition of molecules inside the cell through spectral signals without damaging it.
The scientists tested the method on skin and lung tissues from 2- and 26-month-old mice, combining gene expression and biochemical composition data. Cells in the skin and lungs of older mice showed increased lipid synthesis and accumulation. In senescent skin cells, processes related to muscle contraction, collagen, and extracellular matrix remodeling also changed, while genes associated with immune responses and inflammation became more active in lung tissue.
Machine learning then selected signals most closely associated with senescence from a large number of Raman spectra and gene expression features and combined them into a unique “barcode.” These spectral peaks can indicate changes in lipids, proteins, and other molecules inside the cell.
If the method is confirmed in the future, examining several key sections of a Raman spectrum could help quickly assess whether a cell has entered a state of senescence.
The main difference with this approach is that researchers try to identify a senescent cell not after breaking it down, but through chemical signals obtained while it is alive. This changes how the problem is measured: instead of relying on individual markers such as p16 and p21, machine learning looks for a combination of several features that could provide a more comprehensive picture of the cell’s state.
At the same time, the results of this work are still based on studies of mouse skin and lung tissue, so the research primarily represents a new method rather than an established approach for use in humans. It is also notable that different changes were observed in different tissues during the same study, indicating that cellular senescence may not be expressed through a single universal marker. For future research, a key question will be how stable this “barcode” is across different tissues and conditions.
