基于概率话题模型的轻微型肝性脑病患者肠道菌群结构研究

Study of the structure of intestinal microflora in patients with mild hepatic encephalopathy based on probability topic model

摘要:

[目的]利用LDA（Latent Dirichlet Allocation）概率话题模型分析轻微型肝性脑病（MHE）患者服用利福昔明联合益生菌对其肠道菌群结构异质性和临床疗效的影响。[方法]采用R语言包中的LDA概率话题模型的折叠Gibbs抽样蒙特卡洛算法，对MHE患者肠道菌群结构的时间异质性OTUs（operational taxonomic unit）数据集进行分析。[结果]LDA模型将MHE患者的42份粪便样本分成3个主题（topic），并能鉴定出影响MHE患者肠道菌群异质性结构最大的OTUs菌属，分别为埃希菌属（Escherichia）、类杆菌属（Bacteroides）和链球菌属（Streptococcus）。对比治疗前后，这3种菌属在组内的变异模式为同类型菌属的转变次数和频率均高于不同类型的菌属。利福昔明联合益生菌治疗组和单独利福昔明治疗组治疗后，MHE患者的肠道菌群结构均有所改变（P<0.05）。此外，根据临床疗效指标，对比两组患者治疗后血清IL-2、IL-4、IL-6、IL-10、TNF-α、TBIL、ALT、CRP、NCT-A、γ-GGT及血氨水平，观察组明显优于对照组，差异显著，有统计学意义（P<0.05）。治疗组总有效率88.8%，不良反应总发生率22.2%，对照组总有效率75%，不良反应总发生率38.5%（P<0.05）。[结论]LDA模型不仅能有效地量化菌群结构的异质性，还能鉴定出相对应影响异质性最大的OTUs。利福昔明联合益生菌疗法能明显改善MHE患者的血氨水平和血清炎性因子水平，且对MHE患者的肠道菌群结构也有一定的改变，具体表现为致病菌数量减少，有益菌数量增加，具有较好的临床应用价值。

Abstract:

[Objective] We used Latent Dirichlet Allocation (LDA) probabilistic topic model to analyze the effect of rifaximin combined with probiotics orally taken by patients with mild hepatic encephalopathy (MHE) on structural heterogeneity and clinical efficacy of intestinal microflora. [Methods] We adopted Monte Carlo algorithm for folding Gibbs sampling of LDA probabilistic topic model included in R language to analyze the OTUs (operational taxonomic unit) data set with temporal heterogeneity of MHE patients' intestinal microflora structure. [Results] Using LDA model we divided MHE patients' 42 fecal samples into 3 topics and determined OTUs genus with largest impact on MHE patients' heterostructure of intestinal microflora. They were Escherichia, Bacteroides and Streptococcus. Compared conditions before and after treatment, the variation patterns of 3 species in the group were the genus of the same type. Their variation times and frequency were higher than those of different types of bacteria. Both MHE patients in Rifaximin combined with probiotics treatment group and rifaximin alone treatment group changed the intestinal microflora structure after treatment (P<0.05). In addition, according to the clinical curative effect index:comparing serum IL-2, IL-4, IL-6, IL-10, TNF-α, TBIL, ALT, CRP, NCT-A, γ-GGT and blood ammonia level after treatment, the observation group was obviously superior to the control group with significant differences and the difference has statistical significance (P<0.05). Total effective rate for treatment group was 88.8%, with 22.2% total incidence of adverse reactions; Total effective rate for control group was 75%, with 38.5% total incidence of adverse reactions (P<0.05). [Conclusion] LDA model can effectively quantifying the heterogeneity of flora structure and identify the OTUs with the greatest impact on heterogeneity. Rifaximin combined with probiotics treatment can significantly improve the level of blood ammonia and serum inflammatory factors in patients with MHE. It can also change intestinal microflora structure in MHE patients, which is mainly represented by the reduction of pathogens and the increase of probiotics. As a result, it has good clinical application value.