Machine learning model for predicting age in healthy individuals using age-related gut microbes and urine metabolites

Abstract

Age-related gut microbes and urine metabolites were investigated in 568 healthy individuals using metataxonomics and metabolomics. The richness and evenness of the fecal microbiota significantly increased with age, and the abundance of 16 genera differed between the young and old groups. Additionally, 17 urine metabolites contributed to the differences between the young and old groups. Among the microbes that differed by age, Bacteroides and Prevotella 9 were confirmed to be correlated with some urine metabolites. The machine learning algorithm eXtreme gradient boosting (XGBoost) was shown to produce the best performing age predictors, with a mean absolute error of 5.48 years. The accuracy of the model improved to 4.93 years with the inclusion of urine metabolite data. This study shows that the gut microbiota and urine metabolic profiles can be used to predict the age of healthy individuals with relatively good accuracy.

Keywords: Age; prediction; urine; fecal; metabolomics; metataxonomics.

Figure 1.

Analysis of gut microbial community profiles according to age. The samples were divided into three groups (young: 20–39 years, middle-aged: 40–59 years, and old: ≥60 years). (a) Relative abundance (%) of bacteria by ascending age groups at the phylum level. (b) Principal component analysis (PCA) of beta-diversity based on the operational taxonomic unit (OTU) level (unweighted UniFrac). (c) Alpha-diversities of microbial communities between young (yellow), middle-aged (blue), and old (red) groups (p value was calculated using a Kruskal–Wallis test; *p < .1, **p < .01, ***p < .001). (d, e) Violin plots of relative abundance of bacterial taxa (genus level) that contribute to differences in the linear discriminant analysis effect size (LEfSe) (LDA >2.0 and p < .05) between young and old groups. (d) Relatively higher abundance in the young group. (e) Relatively higher abundance in the old group. p value was calculated using a Kruskal–Wallis test; *p < .1, **p < .01, ***p < .001.

Figure 2.

Orthogonal partial least squares discriminant analysis (OPLS-DA) score plot obtained from gas chromatography mass spectrometry (GC-MS) data of urine. The samples were divided into three groups (young: 20–39 years, middle-aged: 40–59 years, and old: ≥60 years). (a) OPLS-DA score plot of young (yellow), middle-aged (blue), and old (red) groups. (b) OPLS-DA score plot of young (yellow) and old (red) groups. Cross validation was performed using a permutation test that was repeated 200 times. No over-fitting was observed. (c-e) Violin plots of identified metabolites that contribute to differentiation between the young and old groups in the OPLS-DA model (VIP >1.0 and p < .05). (c) Relatively higher intensity in the young group. (d) Relatively higher intensity in the middle-aged group. (e) Relatively higher intensity in the old group. p value was calculated using a Kruskal–Wallis test; *p < .1, **p < .01, ***p < .001.

Figure 3.

Data integration analysis for biomarker discovery using latent components (DIABLO) graphical outputs on the Korean age study. (a) Circos plot of correlations between urine metabolites and fecal microorganisms. Each dataset is assigned a different color: metabolites are in blue blocks and microorganisms in green blocks. Red and blue lines indicate positive and negative correlations between two variables, respectively (r ≥ |0.50|). (b) Network visualization of urine metabolites and fecal microorganisms derived from circos plots. Each dataset is assigned a different color: metabolites are in blue circles and microorganisms in green circles. Red and green lines indicate positive and negative correlations between two variables, respectively.

Figure 4.

Linear regression scatter plot (Pearson correlation) of predicted age versus actual age based on gut microorganisms and urine metabolites (R² = 0.79, p < .001). The x-axis shows the predicted age of the volunteers in years. The y-axis shows the actual age of the volunteers in years. Every blue dot displays one sample. The dotted line shows the linear correlation. R² = coefficient of determination, MAE = mean absolute error in years.

Figure 5.

Sample characteristics and workflow of young, middle-aged, and old groups.