Machine learning predicts lifespan and suggests underlying causes of death in aging C. elegans.

Abstract

Aging leads to age-related pathology that causes death, and genes affect lifespan by determining such pathology. Here we investigate how age-related pathology mediates the effect of genetic and environmental interventions on lifespan in C. elegans by means of a data-driven approach employing machine learning (ML). To this end, extensive data on how diverse determinants of lifespan (sex, nutrition, genotype, mean lifespan range: 7.5 to 40 days) affect patterns of age-related pathology was gathered. This revealed that different life-extending treatments result in distinct patterns of suppression of senescent pathology. By analysing the differential effects on mid-life pathology levels and lifespan, the ML models developed were able to predict lifespan variation, explaining 79% of the variance. Levels of pathology in the pharynx and intestine proved to be the strongest predictors of lifespan. This suggests that elderly C. elegans die predominantly from late-life disease affecting these organs. In addition, we noted profound sex differences in age-related pathology: the striking age-related pathologies in hermaphrodites affecting organs linked to reproduction are absent from males, suggesting that reproductive death may be hermaphrodite limited.