Predicting age from binarized human oral microbial data combined with an ensemble of classifiers

It is well established that the composition of the human microbiome changes with age; however, limited research has explored the association between the oral microbiome and aging, as well as its potential for age prediction. In this study, we investigated the correlation between the oral microbiome and age by analyzing samples from 150 individuals across a wide age range (6-78 years). The observed species richness and Chao1 index significantly increased with age. Permutational multivariate analysis of variance (PERMANOVA), using both Bray-Curtis and Jaccard distances, identified age as a major factor influencing microbial variation. After a comprehensive comparison of four different oral microbiome data processing approaches, we then developed an ensemble model based on binarized oral microbial data, incorporating an eXtreme Gradient Boosting (XGBoost) algorithm with 32 classifiers. This ensemble model achieved a mean absolute error (MAE) of 7.20 years in the independent validation set (n = 15) and 4.33 years in the 20-59 age subgroup (n = 12), significantly outperforming traditional models. When the sample size increased to 2,550, the MAE in the independent validation set (n = 255) was reduced to 4.80 years, with the 20-59 age subgroup (n = 232) achieving an MAE of 3.76 years, highlighting its generalizability and robustness. Additionally, compared to the previously published model for age prediction based on oral microbiome, our model demonstrated significantly superior performance. These findings support the potential of integrating binarized microbial data with ensemble modeling as a promising direction for human age prediction based on the microbiome.IMPORTANCEPERMANOVA analysis with Jaccard distances revealed age as a major determinant of variation in microbial composition, highlighting the potential of binarized oral microbial data as a novel predictor for human age prediction. Furthermore, we developed an ensemble model combining an XGBoost algorithm and 32 classifiers to predict age from binarized oral microbial data. The model achieved an MAE of 7.20 years in the independent validation set (n = 15) and 4.33 years in the 20-59 age subgroup (n = 12). When applied to predict age in 2,550 samples from previous studies, the ensemble model outperformed the prior model, achieving an MAE of 4.44 years compared to the previous model's 4.94 years. These findings demonstrate that binarized oral microbial data, along with the ensemble model we developed, can effectively predict human age and provide a solid foundation for future age-related research.