基于微生物组新颖指数构建龋病菌群诊断模型

Construction of a caries diagnosis model based on microbiome novelty score

目的探讨基于口腔菌群16S核糖体RNA(rRNA)数据的龋病菌群分析与诊断模型的构建及优化。方法检索NCBI、MG-RAST、EMBL-EBI、QIITA等微生物组公开数据库,收集全球范围内人类口腔微生物组的相关研究中所涉及的微生物组数据。通过口腔微生物组搜索引擎(MSE)将龋病测试数据集中的样本(1703例)与健康样本(20540例)进行比对,得到微生物组新颖指数(MNS)并基于该指数构建龋病诊断模型。最后使用非参数多元方差分析比较不同宿主因素对口腔菌群MNS的影响大小,通过控制相关因素优化模型,并利用受试者工作特征(ROC)曲线评价模型效果。结果1)相比于健康样本,龋病样本的微生物多样性降低且菌群结构变异增大;2)ROC曲线对龋病测试数据集进行评估,计算ROC曲线下面积(AUC)为0.67;3)分析表明龋病状态、国家、年龄、龋失补指数(DMFT)及取样位点这5类宿主因素对微生物组MNS有显著影响(P=0.001);4)控制相关宿主因素后的优化诊断模型在中国儿童高龋、中龋、低龋以及混合牙菌斑样本数据集AUC达到0.87、0.74、0.74和0.75。结论基于口腔菌群龋病数据分析构建的龋病诊断模型具有良好的诊断效能。

Objective This study aimed to analyze the bacteria in dental caries and establish an optimized dental-caries diagnosis model based on 16S ribosomal RNA(rRNA)data of oral flora.Methods We searched the public databases of microbiomes including NCBI,MG-RAST,EMBL-EBI,and QIITA and collected data involved in the relevant research on human oral microbiomes worldwide.The samples in the caries dataset(1703)were compared with healthy ones(20540)by using the microbial search engine(MSE)to obtain the microbiome novelty score(MNS)and construct a caries diagnosis model based on this index.Nonparametric multivariate ANOVA was used to analyze and compare the impact of different host factors on the oral flora MNS,and the model was optimized by controlling related factors.Finally,the effect of the model was evaluated by receiver operating characteristic(ROC)curve analysis.Results 1)The oral microbiota distribution obviously differed among people with various oral-health statuses,and the species richness and species diversity index decreased.2)ROC curve was used to evaluate the caries data set,and the area under ROC curve was AUC=0.67.3)Among the five hosts’factors including caries status,country,age,decayed missing filled tooth(DMFT)indices,and sampling site displayed the strongest effect on MNS of samples(P=0.001).4)The AUC of the model was 0.87,0.74,0.74,and 0.75 in high caries,medium caries,low caries samples in Chinese children,and mixed dental plaque samples after controlling host factors,respectively.Conclusion The model based on the analysis of 16S rRNA data of oral flora had good diagnostic efficiency.