Machine Learning Screens For Sepsis Signature Predictor Genes

Background and Purpose:Sepsis places a huge burden on the healthcare system,with millions of people dying each year from sepsis,which can be further improved by early diagnosis and timely medical treatment before the disease becomes serious.Sepsis has become a major global public health challenge.The pathogenesis of sepsis and its complexity are not only the process of systemic inflammatory response or immune disorder,but also the disorder of multiple organ functions of the body.At the cellular and molecular levels,sepsis involves mechanisms such as dysregulated inflammatory response,immunosuppression,apoptosis,and endoplasmic reticulum stress.The complex pathophysiology of sepsis leads to the release of many biomarkers.By measuring more biomarkers,the host response to infection can be better assessed,which can better guide clinical treatment.The emergence and development of bioinformatics has provided new ideas for finding the pathogenesis,biomarkers and therapeutic targets of sepsis.This study aims to find sepsis feature prediction genes by statistical method LASSO and machine learning method SVM-RFE method,and explore the application of machine learning in screening sepsis feature prediction genes.Methods and Results:1.We downloaded the expression profiles of three sepsis-related whole blood genes of GSE26440,GSE54514 and GSE57065 from the GEO database.2.The datasets of GSE26440 and GSE57065 were used as the experimental groups for feature gene screening,a total of 19 sepsis characteristic genes were screened by the Lasso algorithm,13 sepsis characteristic genes were screened by the SVM algorithm,and 5 sepsis candidate characteristic genes were identified by intersecting the Lasso results with the sepsis characteristic genes screened by SVM.3.GSE54514 was used as the validation group to verify the five sepsis screening characteristic genes,and five DEm RNAs of MCEMP1,UPP1,CD177,CYSTM1 and RAB13 were identified as diagnostic gene biomarkers.4.GO annotation analysis of differential genes revealed that the differential genes involved inflammatory response included T-cell activation,immune response regulation signaling pathway,leukocyte-mediated immunity,activation of immune response,signaling pathway of immune response regulating cell surface receptors,cell-cell aggregation of white blood cells,signaling pathways of immune response activation of cell surface receptors,signal transduction of immune response activation,T-cell differentiation,etc.KEGG analysis of differential genes showed that5.The immune cell infiltration analysis using CIBERSORT for differential genes showed that neutrophils,monocytes and M0 macrophages had significantly higher expression levels in sepsis blood,CD177,CYSTM1,MCEMP1,RAB13,UPP1 were associated with increased expression levels of M0 macrophages,and CD177,CYSTM1,MCEMP1,RAB13,UPP1 were associated with elevated expression levels of neutrophils.6.A total of five sepsis characteristic prediction genes of MCEMP1,UPP1,CD177,CYSTM1 and RAB13 were identified.In conclusion,the statistical method LESO and the machine learning method SVM-RFE method are used to screen the characteristics of sepsis for biomarkers,and provide an early interpretation of sepsis co-screening method.MCEMP1,UPP1,CD177,CYSTM1 and RAB13 were screened as sepsis characteristic prediction genes,which may help the early diagnosis and prognosis of sepsis.