基于微生物组学和机器学习构建咽喉反流唾液菌群诊断模型

The Salivary Microbiota Diagnostic Model for Laryngopharyngeal Reflux Based on Microbiome and Machine Learning

目的探索使用微生物组学技术和机器学习的方法构建咽喉反流(laryngopharyngeal reflux,LPR)唾液菌群诊断模型的可能性。方法以2022年2~11月就诊的经8周质子泵抑制剂治疗显效的34例LPR患者为病例组,与病例组年龄、性别和体重指数相匹配的47例健康志愿者为对照组,留取用药前的唾液标本,提取DNA,扩增16S片段后进行测序,对测序结果进行生物学信息分析,比较菌属水平物种差异。选择病例组24例、对照组33例作为训练集,其余作为测试集,随机森林法分类数据,进行十折交叉验证,选择最优菌属组合构建诊断模型,计算疾病概率指数(probability of disease,POD),绘制ROC曲线,使用SPSS 18.0软件对数据进行统计学分析。结果病例组和对照组相比,唾液中22个属相对丰度存在统计学差异(P<0.05),构建由6个菌属组成的诊断模型,即乳酸杆菌属、芽孢杆菌属、新鞘氨醇杆菌属、假交替单胞菌属、罗尔斯通菌属和Phocaeicola属,测试集ROC曲线下面积为0.843,诊断模型灵敏度为60.0%,特异度为87.71%,Kappa值为0.470。结论由唾液菌群构建的菌属组合诊断模型可以区分LPR患者与健康人,具有潜在的临床应用价值。

Objective To study the possibility of salivary microbiota model to diagnose laryngopharyngeal reflux(LPR).Methods A case-control study was applied to enroll 34 patients as case group who showed significant efficacy after 8 weeks of proton pump inhibitor treatment from February 2022 to November 2022.And 47 healthy volunteers matched by age,gender and body mass index with the case group were enrolled as the control group.Their salivary samples were collected before medication,and the salivary microbiota was detected by 16S rDNA sequencing.Bioinformatics analysis was conducted on the sequencing results to compare species differences at the genus level.A total of 24 patients and 33 cases in the control group were selected as train set and the rest as test set.Random forest method was used to classify data and ten fold cross validation was applied to select the optimal bacterial genus combination to construct a diagnostic model.The probability of disease(POD)index was calculated and receiver operating characteristic curve(ROC)was used to evaluate the diagnostic model in diagnosis of LPR.SPSS 18.0 software was utilized for statistical analysis.Results Compared with the control group,there was a statistical difference in the relative abundance of 22 genera in saliva between the case group and the control group(P<0.05).A diagnostic model consisting of 6 genera was constructed,namely Lactobacillus,Novosphingobium,Bacillus,Pseudoalteromonas,Ralstonia and Phocaeicola.The area under the ROC curve of the test set was 0.843,the sensitivity of the diagnostic model was 60.0%,the specificity was 87.71%,and the Kappa value was 0.470.Conclusion The bacterial combination diagnostic model constructed from saliva microbiota based on microbiome and machine learning can effectively distinguish LPR patients from healthy individuals,which has potential clinical application value.