Support vector machine-based differentiation between aggressive and chronic periodontitis using microbial profiles
Support vector machine-based differentiation between aggressive and chronic periodontitis using microbial profiles
复制标题
DOI：
10.1111/idj.12326
复制
发表时间：
2018-02-01
影响因子：
3.3
通讯作者：
Levin, Liran
中科院分区：
文献类型：
--
作者：
Feres, Magda;Louzoun, Yoram;Levin, Liran
背景：不同牙周状况下是否存在特定的微生物谱仍然是一个有争议的问题。本研究的目的是测试40种细菌可用于利用机器学习将患者分类为全身性慢性牙周炎（ChP），全身性侵袭性牙周炎（AgP）和牙周健康（PH）的假设。方法：收集AgP、ChP和PH患者的龈下生物膜样品，采用棋盘DNA-DNA杂交法分析其40种细菌的含量。然后进行了两个阶段的机器学习。首先，我们测试了PH和疾病中的细菌群落组成是否存在差异，然后我们测试了ChP和AgP之间的细菌群落组成是否存在差异。在每次分析中将数据分为70%训练和30%测试。支持向量机（SVM）分类器使用线性内核和框约束为1。分析分为两个部分。结果：总体而言，435例患者（3，915份样本）纳入分析（PH = 53; ChP = 308; AgP = 74）。在所有的主成分分析（PCA）方向的健康样本的方差小于牙周病的样本，这表明PH的特点是由一个统一的细菌组成和牙周病的样本的细菌组成是更加多样化。相对细菌负荷可以区分AgP和ChP。结论：SVC分类器使用一组40种细菌能够区分PH，AgP在年轻人和ChP。
Background: The existence of specific microbial profiles for different periodontal conditions is still a matter of debate. The aim of this study was to test the hypothesis that 40 bacterial species could be used to classify patients, utilising machine learning, into generalised chronic periodontitis (ChP), generalised aggressive periodontitis (AgP) and periodontal health (PH). Method: Subgingival biofilm samples were collected from patients with AgP, ChP and PH and analysed for their content of 40 bacterial species using checkerboard DNA-DNA hybridisation. Two stages of machine learning were then performed. First of all, we tested whether there was a difference between the composition of bacterial communities in PH and in disease, and then we tested whether a difference existed in the composition of bacterial communities between ChP and AgP. The data were split in each analysis to 70% train and 30% test. A support vector machine (SVM) classifier was used with a linear kernel and a Box constraint of 1. The analysis was divided into two parts. Results: Overall, 435 patients (3,915 samples) were included in the analysis (PH = 53; ChP = 308; AgP = 74). The variance of the healthy samples in all principal component analysis (PCA) directions was smaller than that of the periodontally diseased samples, suggesting that PH is characterised by a uniform bacterial composition and that the bacterial composition of periodontally diseased samples is much more diverse. The relative bacterial load could distinguish between AgP and ChP. Conclusion: An SVC classifier using a panel of 40 bacterial species was able to distinguish between PH, AgP in young individuals and ChP.