細胞老化を逆転させる遺伝的手がかりの高速探索Fast-tracking genetic leads to reverse cellular aging

Google DeepMind has published research describing a computational approach to rapidly pinpoint the genetic factors that can reverse cellular aging, a step the company frames as a way to accelerate the discovery of treatments for age-related disease. The work sits at the intersection of machine learning and genomics, and it matters because cellular aging underlies a broad range of conditions, from neurodegeneration to cardiovascular decline, that grow more common as populations age.

At the center of the research is the challenge of scale. The human genome contains roughly 20,000 protein-coding genes, and the number of possible combinations of genetic interventions that might influence aging is effectively astronomical. Testing each candidate experimentally in the laboratory is slow and costly, so researchers can realistically explore only a small fraction of the search space. DeepMind's approach appears to use models that predict which genetic changes are most likely to shift cells from an aged state toward a more youthful one, allowing scientists to prioritize the most promising candidates for wet-lab validation rather than screening blindly.

A key concept underpinning this line of work is cellular reprogramming. In 2006, the Japanese scientist Shinya Yamanaka showed that introducing a small set of transcription factors, now known as the Yamanaka factors, could convert mature cells back into a pluripotent, stem-cell-like state. Subsequent research demonstrated that partial reprogramming can reverse some markers of aging without erasing a cell's identity entirely. Identifying which genes to target, and how to modulate them safely, has remained a central bottleneck. Machine learning is increasingly being applied to narrow that search, learning patterns from large datasets of gene expression and cellular behavior to suggest interventions that a human researcher might not prioritize.

The reported method is likely to combine several data types, including transcriptomic profiles that capture which genes are active in young versus old cells, and models trained to recognize the molecular signatures associated with biological age. So-called aging clocks, which estimate a cell or tissue's biological age from epigenetic or expression data, provide one way to measure whether an intervention is moving cells in the desired direction. By pairing predictive models with these readouts, the system can, in principle, rank candidate genes and generate hypotheses far faster than traditional trial-and-error screening.

This research fits into a broader pattern of DeepMind applying its expertise to biology. The company's AlphaFold system, which predicts three-dimensional protein structures from amino acid sequences, reshaped structural biology and contributed to its founders receiving a share of the 2024 Nobel Prize in Chemistry. DeepMind has since extended that work through AlphaFold 3, which models interactions between proteins and other molecules, and through Isomorphic Labs, a sister company focused on AI-driven drug discovery. Aging research represents an adjacent frontier where the same combination of large datasets and predictive modeling could shorten discovery timelines.

The field also includes well-funded dedicated efforts. Altos Labs, backed by billions of dollars in investment, and Calico, an Alphabet-affiliated company launched in 2013, are both pursuing the biology of aging and cellular rejuvenation. Academic groups continue to study the hallmarks of aging, a framework describing the molecular and cellular processes, such as genomic instability and cellular senescence, thought to drive the aging phenotype. DeepMind's contribution appears aimed less at replacing this experimental science than at guiding it toward the interventions most worth testing.

Important caveats remain. Computational predictions must be confirmed in living cells and, eventually, in animal models before any therapeutic relevance can be established, and results in cultured cells do not always translate to whole organisms. Reprogramming approaches also carry safety concerns, including the risk of tumor formation if cells lose their identity or divide uncontrollably. Any path from a genetic hypothesis to an approved therapy would span many years of validation and clinical testing, and the current work is best understood as an early-stage research tool rather than a treatment.

Even so, the announcement reflects a growing conviction that artificial intelligence can compress the exploratory phase of biological discovery. If the models reliably surface genuine reversers of cellular aging, they could help researchers focus limited laboratory resources on the highest-value experiments, a benefit that would extend well beyond longevity into any area where the space of possible interventions dwarfs the capacity to test them.