With neurodegenerative diseases on the rise, findings ways to slow brain aging is a priority for many researchers. (Image by meeboonstudio on Shutterstock)
In a Nutshell
- What they did: Scientists created an AI-driven “brain aging clock” using gene expression data from 778 human brain donors aged 20 to 97. The system identified patterns that predict biological brain age and screened over 43,000 chemical and genetic perturbation profiles to find compounds that could reverse age-related changes in brain cells.
- What they found: The AI identified 453 distinct compounds predicted to make aging brain cells appear more youthful. Some had already been shown to extend lifespan in animals. A three-drug combination selected from the predictions was tested in aged mice, where it reduced anxiety-like behavior and partially restored youthful gene expression patterns in brain tissue. Memory improvements were observed but did not reach statistical significance.
- Why it matters: This study offers a new method to rapidly identify potential brain rejuvenation therapies using transcriptomic data and machine learning. With neurodegenerative diseases on the rise, the approach could accelerate the search for treatments that preserve brain function and slow cognitive aging.
DERIO, Spain — What if doctors could determine how old someone’s brain actually is — not based on their birth certificate, but on how youthful their neural tissue appears at the cellular level? A new study has created an artificial intelligence system that does exactly that, successfully identifying hundreds of potential treatments that may help reverse aspects of brain aging.
The research, published in Advanced Science, introduces a machine-learning platform that pinpointed 453 unique compounds predicted to make aging brain cells appear more youthful. When researchers tested a combination of three such compounds in older mice, they observed reduced anxiety, signs of molecular rejuvenation in brain tissue, and a trend toward improved memory, though the improvement was not statistically significant.
How AI Learned to Detect Brain Aging Patterns
Researchers from the Luxembourg Centre for Systems Biomedicine and collaborating institutions trained their AI using brain tissue samples from 778 healthy individuals, ages 20 to 97. The model learned to recognize age-related shifts in gene expression. Essentially, it found patterns indicating which genes are more or less active as the brain gets older.
The system ultimately identified 365 key genes whose activity correlates strongly with aging. Much like a doctor assessing skin health by looking at wrinkles, this tool assesses brain aging by examining subtle molecular changes in gene activity.
Its accuracy was impressive: it predicted chronological age within about five years based on gene expression alone. When analyzing tissue from people with neurodegenerative diseases, the predicted brain age was often far older than the person’s actual age, reinforcing the link between neurodegeneration and accelerated aging.
Database Analysis Reveals Potential Youth-Restoring Drugs
With the AI “aging clock” in place, researchers analyzed a massive dataset of 43,840 chemical and genetic perturbation profiles in brain cells. They were looking for interventions that could shift gene expression patterns toward a more youthful state.
The system identified 971 rejuvenating perturbations in neural progenitor cells and 68 in mature neurons. From these, 453 distinct compounds emerged. Some had previously extended lifespan in animal models, supporting the validity of the approach.
The list included a wide range of substances, including experimental compounds and drugs already approved for other uses. This diversity suggests that brain aging may be addressed through multiple biological pathways; although further research is needed to clarify how these treatments work and which are most effective.
Mouse Studies Show Reduced Anxiety and Molecular Rejuvenation
To test whether the predictions translated into real-world effects, researchers selected three compounds: 5-azacytidine, tranylcypromine, and JNK-IN-8.
They treated aged mice with this three-drug combination for four weeks. Treated mice spent more time in open areas during anxiety tests, a behavior indicating reduced fearfulness. Spatial memory tests showed a modest trend toward improvement, although the difference was not statistically significant.
The most striking effects appeared at the molecular level. Treated mice showed gene expression patterns in the brain that more closely resembled those of much younger animals — a partial “rejuvenation” of the transcriptome.
Clinical Applications and Future Research Directions
Interestingly, the 365 genes used by the AI to track brain aging weren’t limited to brain-specific functions. Many were involved in fundamental processes like DNA repair, RNA transcription, and chromatin dynamics, pointing to shared mechanisms of aging across tissues.
The researchers also found that people with neurodegenerative diseases in their 60s and 70s had brain aging signatures about 15 years older than their actual age. This age gap narrowed in older individuals, suggesting that early intervention might offer the greatest benefit.
Several of the AI-identified compounds are already approved for other conditions, potentially paving a faster path toward clinical testing in aging-related cognitive decline. The researchers released their system as an open-source R package called brainAgeShiftR, enabling other scientists to test and build on their work.
As the global population ages — with over two billion people projected to be over 60 by 2050 — the need for effective strategies to preserve brain health grows more urgent. This AI-based platform could open new doors in the search for treatments that help the aging brain stay sharper for longer.
Disclaimer: This report is based on a preclinical study using mouse models and computational screening. The findings have not yet been tested in humans, and the long-term safety or effectiveness of the identified compounds remains unknown. Interpretations of AI-predicted treatments should be viewed as exploratory and require further validation.
Paper Summary
Methodology
Researchers developed a brain-specific transcriptomic aging clock by analyzing gene expression data from 2,456 postmortem brain tissue samples taken from 778 healthy individuals aged 20 to 97. The clock, based on the expression of 365 genes, was trained to estimate biological brain age with high accuracy. Using this model, the team screened 43,840 transcriptional profiles representing 5,771 chemical and genetic perturbations applied to neural progenitor cells and neurons. From this screening, the system identified 453 unique compounds predicted to reduce transcriptional age. For in vivo validation, the researchers selected three of these compounds—5-azacytidine, tranylcypromine, and JNK-IN-8—and administered them as a combination treatment to aged mice over four weeks. Behavioral and transcriptomic analyses followed.
Results
The transcriptomic aging clock predicted chronological brain age within an average margin of 4 to 6 years and demonstrated strong performance across different brain regions, sexes, and external validation datasets. The model identified 971 rejuvenating perturbations in progenitor cells and 68 in neurons, amounting to 453 distinct compounds with age-reversing potential. Some of these predicted compounds had previously extended lifespan in animal studies, adding credibility to the AI’s outputs. When the three-drug combination was tested in aged mice, the animals displayed reduced anxiety-like behavior and showed gene expression patterns in the brain that resembled those of younger mice. Spatial memory performance showed a trend toward improvement, although the difference was not statistically significant. Enrichment analysis further confirmed that the gene expression changes induced by treatment overlapped significantly with those typically seen in younger animals.
Limitations
The study relied on bulk brain tissue samples rather than cell-type-specific data, which may limit the resolution of age-related changes in specific neuronal populations. The perturbation data used for screening were limited to just two cell types—neurons and neural progenitor cells—potentially narrowing the generalizability of the findings. Additionally, only one combination of three compounds was tested in animals, so the individual contribution of each drug remains unknown. The machine learning model assumes a linear relationship between gene expression and age, which may not fully capture the complexity of aging biology. Finally, the long-term safety and side effects of the compound combination were not assessed in this study.
Funding and Disclosures
The study was supported by the Luxembourg Centre for Systems Biomedicine, CIC bioGUNE in Spain, IKERBASQUE (the Basque Foundation for Science), and the University of Santiago de Compostela. While the paper lists multiple international collaborators, full funding and disclosure information was not provided in the available excerpts and should be confirmed by reviewing the complete publication.
Publication Details
This research was published in Advanced Science in 2025 under the title “A Machine-Learning Approach Identifies Rejuvenating Interventions in the Human Brain.” The authors include Guillem Santamaria, Cristina Iglesias, Sascha Jung, Javier Arcos Hodar, Ruben Nogueiras, and Antonio del Sol, with affiliations spanning Luxembourg, Spain, and the Basque Country. The work represents a collaboration between several European institutions specializing in systems biomedicine and aging research. The open-access paper is available via Wiley with DOI: 10.1002/advs.20250334
