Researchers analyzed skin cells from the very young to the very old and looked for molecular signatures that can be predictive of age. By applying machine-learning algorithms to these biomarkers, they were able to predict a person's actual age with less than eight years error, on average.