New models predict biological age using blood and microbiome

A new research paper was published in Volume 18 of Aging-US on March 12, 2026, titled "Blood biochemical and gut microbiotic neural network models forecasting human biological age."

Led by Anastasia A. Kobelyatskaya from the Russian Clinical Research Center for Gerontology, Pirogov Russian National Research Medical University, and the Institute of Biology of Aging and Healthy Longevity Medicine with Preventive Medicine Clinic, Petrovsky Russian Research Centre of Surgery - with corresponding author Alexey Moskalev from the Institute of Biology of Aging and Healthy Longevity Medicine with Preventive Medicine Clinic, Petrovsky Russian Research Centre of Surgery - the study builds a gender-specific biochemical model (seven routine clinical markers, e.g., cystatin-C, IGF-1, DHEAS, plus sex-specific sets) and a microbiota model (45 species measured by full-length 16S sequencing). Both models were trained and tested on the same 637-person dataset and achieved mean absolute errors of around six years and R² values above 0.8.

The team emphasised interpretability: they applied SHapley Additive exPlanations (SHAP) to convert each model from a "black box" into a more interpretable tool, showing how individual predictors (for example, DHEAS, cystatin-C, NT-proBNP in the blood model, and species such as Blautia obeum in the microbiota model) shift predicted age in years for a given individual. The biochemical clock yielded a small (clinically accessible) predictor set (7 markers) to ease clinical translation, while the microbiota clock used a 45-species signature and highlighted microbiome taxa whose abundance gradients correlate with predicted microbiotic age.

"As the proposed models possess both global and local explainability, they hold future potential for application in monitoring the effectiveness of various interventions in clinical trials."

The authors note limitations and next steps: the cohort was restricted to a Caucasian population, and the microbiota model requires sequencing resources that may limit immediate clinical rollout. They call for external validation in larger, ethnically diverse cohorts, prospective testing to link model predictions to health outcomes, and application of the explainable models to monitor responses in intervention trials (for example, lifestyle, diet, or drug studies) where a change in predicted biological age would be an early, interpretable signal of benefit.