Novel Clustering Methods Identified Three Caries Status-Related Clusters Based on Oral Microbiome in Thai Mother-Child Dyads

Abstract

Early childhood caries (ECC) is a disease that globally affects pre-school children. It is important to identify both protective and risk factors associated with this disease. This paper examined a set of saliva samples of Thai mother-child dyads and aimed to analyze how the maternal factors and oral microbiome of the dyads influence the development of ECC. However, heterogeneous latent subpopulations may exist that have different characteristics in terms of caries development. Therefore, we introduce a novel method to cluster the correlated outcomes of dependent observations while selecting influential independent variables to unearth latent groupings within this dataset and reveal their association in each group. This paper describes the discovery of three heterogeneous clusters in the dataset, each with its own unique mother-child outcome trend, as well as identifying several microbial factors that contribute to ECC. Significantly, the three identified clusters represent three typical clinical conditions in which mother-child dyads have typical (cluster 1), high-low (cluster 2), and low-high caries experiences (cluster 3) compared to the overall trend of mother-child caries status. Intriguingly, the variables identified as the driving attributes of each cluster, including specific taxa, have the potential to be used in the future as caries preventive measures.

Keywords: clustering; early childhood caries; maternal influence; oral microbiome; saliva.

Figure 1

Salivary l microbiome profile of the three clusters identified among Thai mother–child dyads. (A) α diversity of salivary microbiome among mothers and children. Microbial variation measured by α diversity Shannon index. t-test was used for statistical analysis. (B) Relative abundance of top 20 genera among mother and children. (C–E) β diversity of salivary microbiome among three clusters of mothers and children. Principle coordinate analysis (PCOA) plot is generated using OTU metrics based on Bray–Curtis index. Permutational MANOVA (PERMANOVA) was used for statistical analysis. (F–H) Taxa at genus level differently enriched in the saliva of mother and children. Linear discriminant analysis (LDA) effect size method was performed to compare taxa between children and mothers. The bar plot lists the significantly differential taxa based on effect size [LDA score (log10) > 2 and FDR < 0.05].

Figure 2

Significantly different maternal factors identified between the three clusters. (A–H) Bar plots of the proportion of subjects within each cluster that indicated yes to the labeled maternal factor of interest. The ones shown above are the ones that are significantly different between the clusters. (I,J) Proportions of each level of the indicated maternal factor within each cluster.

Figure 3

Histograms of significantly different selected OTUs across the three clusters. (A–H) The density of the abundance each OTU within each cluster is represented in a layered fashion to visualize the differences between them.