# Metagenomic research on the structural difference of plaque microbiome from different caries stages and the construction of a caries diagnostic model

## TL;DR

This study explores how the plaque microbiome changes with different stages of dental caries and builds a highly accurate diagnostic model using deep learning.

## Contribution

The study provides high-resolution insights into microbiome shifts across caries stages and introduces a deep learning diagnostic model with over 98% accuracy.

## Key findings

- The CH group had the highest species richness, while the DC group had the lowest.
- The microbial structure showed the greatest difference between CH and DC groups.
- A deep learning model was developed to diagnose caries stages with an AUC of over 98%.

## Abstract

Development of dental caries is a dynamic process; yet, there is limited knowledge on microbial differences at various stages of caries at higher resolution. To investigate the shifting microbiome profiles across different caries stages, 30 children were enrolled in this study, including 15 caries-active patients and 15 caries-free individuals. Plaque samples were collected from the buccal surface of caries-free subjects, defined as confident health (CH; n = 15). For caries-active individuals, plaque samples were collected from non-cavitated surfaces (defined as relative health [RH], n = 15), enamel caries (EC; n = 15), and dentin caries samples (DC; n = 15). All the above samples were sequenced through the 2bRAD sequencing platform to reveal the microbial community structures in each group. We identified significant differences in microbial community structures from different caries stages. First, the CH group showed the highest species richness (P < 0.05), and then followed by the RH and EC groups with lower richness, and the lowest richness was found in the DC group, yet no significant difference was found among the last three groups (P > 0.05). Second, the microbial structure exhibited the greatest difference between CH and DC groups, followed by the distance between RH/EC and DC groups, and the smallest difference was found between RH and EC groups. Third, specific species were found with a significant difference during the different caries stages. Therefore, we developed a diagnostic model using deep learning methods based on neural networks to diagnose different caries stages with an AUC of over 98%. This may provide methodological support for the understanding of the etiological factor in the pathological progression of dental caries.

The diagnosis and treatment of dental caries are crucial for human oral health. Previous studies have focused on the microbial differences between caries and healthy teeth, but there was not enough knowledge on the microbial differences at different stages of dental caries. Our findings could provide a high-resolution understanding of the microbial divergencies among different stages of dental caries and thus build microbial-based diagnostic models for differentiating dental caries status using deep learning methods with an accuracy of over 98%. This may provide methodological support for the understanding of the etiological factor in the pathological progression of dental caries.