The Correlation Between Gut Microbiome And Common Disease And The Establishment Of Constipation Prediction Model

The gut microbiome was closely related to human health and played a very important role in the metabolism and immune system.The rapid development of the next-generation sequencing and bioinformatics technology made the study which focuses on the interaction between the gut microbiome and hosts deeper.However,most of the previous studies based on a single cohort or a relatively small database,which limited the understanding of targeted regulation of gut microbiome on host healthy.Based on the American Gut Project(AGP)and the SRA database,at first,this study performed a series of correlation analysis to explore the potential correlation between hosts background information and the gut microbiome.Then,this studies performed diversity analysis,taxa difference and pathways analysis to get the characters of the gut microbiome of constipation,diabetes,cardiovascular disease,Inflammatory Bowel Disease,irritable bowel syndrome,autism spectrum disorder and obese population.Finally,high machine learning models were constructed by multiple machine learning model algorithms and feature selection method and verified the accuracy by population experiments.All the work mentioned above could provide the theories basis of analyzing the mechanism of the gut microbiome in regulating the health of the host and developing food microbial resources targeted by the gut microbiome.The main findings are as follows:(1)The Adonis,Anosim and MRPP analysis were performed based on Unweighted Unifrac,Weighted Unifrac and Bray-Curts distance matrix to explore the potential correlation between 85 hosts background information,including latitude and longitude,nationality,age,body mass index,bowel frequency,diabetes,and the gut microbiome.The result suggested the geography,body mass index,defecation quality,inflammatory bowel disease,and diabetes are relatively closely related to the gut microbiome.(2)Exploring the characters of the gut microbiome of 7 common diseases,including diabetes,cardiovascular and cerebrovascular diseases,inflammatory enteritis(IBD),irritable bowel syndrome(IBS),obesity,autistic spectrum disorder(ASD)and constipation.Except the constipation,the a-diversity indexes of other disease significantly decreased comparing with normal population,the a-diversity indexes of constipation showed increased trend.The principal component analysis(PCA)based Unweighted Unifrac,Weighted Unifrac and Bray-Curtis did not showed any difference between disease and normal population.After the taxa different analysis,we found all the disease had some significantly different taxa,however the pathway different analysis only found that the gut microbiome of obese,IBD,ASD and constipation had some significantly different pathways.Besides,we constructed four machine learning model to verified the correlation between disease and gut microbiome.(3)First,we constructed a big gut microbiome databased based on AGP and SRA gut microbiome database.Then,to explore the relationship between sample size and model performance,we used the learning curve to analysis their relationship.Finaly,based on the machine learning methods,including the k-nearest neighbour(kNN),support vector machine(S VM),decision tree(DT),Random Forest(RF),Gradient Boost Regression Tree(GBRT),Adaptive Boosting(AdA),Naive Bayes(NB),logistic regression and Lasso model,and feature selection methods,including including e Wilcox rank-sum test,T-test,Mann-whity test,chi2 analysis,F test,mutual information,Logistic regression,Lasso regression,and random forest,we constructed some prediction models to preciton the constipaiton status.The AUC value and the accuracy of the test set and the verification set of the Gradient Boost Regression Tree model after chi2 selection were 87.6%,85.3%and 88.1%respectively.The AUC value and the accuracy of the test set and the verification set of the Gradient Boost Regression Tree model after log selection were 86.2%,81.7%and 86.9%respectively.Both of the two models were the best model in this study.