Share this post:
The human microbiome consists of a community of trillions of micro-organisms, such as bacteria, fungi, viruses, and live all over the body including on the skin, in the mouth and along the digestive tract. A balanced microbiome is important for an individual’s health and wellbeing, including proper functionality the digestive and immune systems.
The human microbiome is constantly evolving and has been observed to change with age. The presence of unusually early microbiome aging patterns, relative to chronological age, could potentially signal altered susceptibility for age-related diseases. Conversely, a “young” microbiome might offer clues on how to decelerate the aging process1. This is why researchers from IBM and University of California San Diego (UC San Diego) recently investigated the robustness of the human microbiome as an indicator of age. The results revealed that the skin microbiome provides the most accurate prediction of age, on average to within a few years of the true age in healthy subjects, compared to oral and gut microbiomes. The research, which is part of an ongoing collaboration on AI for Human Microbiome between IBM Research and UC San Diego under the AI Horizons Network was published today in the journal mSystems, titled: “Human skin, oral, and gut microbiomes predict chronological age.”
In this study, microbiome sequencing data contained in public repositories, including those from fecal (representing gut microbiota), saliva (representing oral microbiota) and skin samples from several continents, was processed with a state-of-the-art bioinformatic pipeline to ensure coherence and compatibility of the heterogeneous data. Machine learning was then applied to predict age from the relative abundance of microbes within the sampled microbiomes. Random forest regression models, tuned, trained and tested for the task, recaptured the previously association between the gut microbiome and age.
The distinct capability for age prediction from gut (A), oral (B), and skin (C) microbiomes. Spline fit to the data is also shown (blue curve). Although the skewed age distribution in the skin or oral microbiota data set may decrease the accuracy of age prediction for the older adults, it will not affect the conclusions about the relative abilities of different human microbiomes to predict age. Prediction performances at increasing numbers of microbial species were obtained by retraining the random forest classifier on the top-ranking features (ASVs), shown in terms of mean absolute error (MAE) from gut (D), oral (E), and skin (F) microbiota identified with previous random forest models trained in different cohorts. Data are from Qiita studies 11757, 10317, 550, 1841, 1774, 2010, 2024, 2202, 11052, and 10052.
However, it was surprising to discover that the skin and oral microbiomes are much more predictive of age than gut microbiome. Promisingly, the hand and forehead skin microbiome age models generalized across cohorts and geographies, indicating studies from several sources could be combined in the future to potentially accelerate discovery from globally available microbiome data.
The gut and oral microbes enriched in young subjects were found to be more abundant and more prevalent than microbes enriched in the old subjects, suggesting a model where aging occurs in tandem with the loss of key microbes over a lifetime. This observation sets the stage for future research on the role of the microbiome in the aging process. Taken together, the results demonstrate that accurate and generalizable indicators of age can be derived from using machine learning on microbiome data.
This work opens the opportunities for developing, with the help of AI, non-invasive microbiome-based diagnostics and also interventions to maintain a healthy “youthful” microbiome.
1: Huang S, Haiminen N, Carrieri A-P, Hu R, Jiang L, Parida L, Russell B, Allaband C, Zarrinpar A, Vázquez-Baeza Y, Belda-Ferre P, Zhou H, Kim H-C, Swafford AD, Knight R, Xu ZZ. 2020. Human skin, oral, and gut microbiomes predict chronological age. mSystems 5:e00630-19. https://doi.org/10.1128/mSystems.00630-19.