Discrimination of normal from slow-aging mice by plasma metabolomic and proteomic features

Tests that can predict whether a drug is likely to extend mouse lifespan could speed up the search for anti-aging drugs. We have applied a machine learning algorithm, XGBoost regression, to seek sets of plasma metabolites (n = 12,000) and peptides (n = 17,000) that can discriminate control mice from mice treated with one of five anti-aging interventions (n = 278 mice). When the model is trained on any four of these five interventions, it predicts significantly higher lifespan extension in mice exposed to the intervention which was not included in the training set. Plasma peptide data sets also succeed at this task. Models trained on drug-treated normal mice also discriminate long-lived mutant mice from their respective controls, and models trained on males can discriminate drug-treated from control females. Triglycerides are over-represented among the most influential features in the regression models. Triglycerides with longer fatty acid chains tend to be higher in the slow-aging mice, while triglycerides with shorter fatty acid chains tend to decrease. Plasma metabolite patterns may help to select the most promising anti-aging drugs in mice or in humans and may give new leads into physiological and enzymatic targets relevant to the discovery of new anti-aging drugs.