关键词: 主成分分析, Stacking模型, 肠道菌群, 脑卒中后抑郁, 菌群多样性

Abstract: Objective In this study,patients with stroke were selected as the research object,and the microbial community diversity of patients' stool samples was sequenced by the second-generation Illumina high-throughput sequencing platform． Twenty four phylum species with 30% species abundance were selected as indicators for the study of gut microbiota,and then the correlation between gut microbiota and post-stroke depression（PSD） was studied．Methods Taking 24 categories of 40 stroke patients as characteristic variables,depression group and control group as dichotomous target variables,a stacking classification model based on Logistic regression,random forest,support vector machine and AdaBoost was established．As the feature selection method of the model,principal component analysis selects the appropriate principal components for model training,and evaluates its performance through dichotomous evaluation reports（precision,recall,f1 score）,ROC curve and confusion matrix．Results The baseline of the two groups（depression group and control group）was consistent（P<0.05）through the difference test．From the perspective of stacking model fusion,the specific intestinal flora affecting post-stroke depression was quantitatively analyzed．The results showed that Actinobacteria,Bacteroidetes,Proteobacteria and Acidobacteria were significantly increased in PSD patients（P<0.001）,while Firmicutes,Verrucomicrobia,Chloroflexi and Tenericutes were significantly decreased in PSD patients（P<0.001）．Conclusions The above microbiota are the main factors affecting the mood of patients with post-stroke depression．Therefore,in clinical practice,we can adjust the depression level of patients with post-stroke depression by properly intervening the changes of intestinal microbiota,which provides a scientific basis for the diagnosis and treatment of PSD．

Key words: principal component analysis, Stacking model, gut microbiota, post-stroke depression, flora diversity