Classifier for identification of robust sepsis subtypes

Abstract

This disclosure provides a gene expression-based method for determining whether a subject having sepsis has an Inflammopathic phenotype, an Adaptive phenotype or a Coagulopathic phenotype. A kit for performing the method is also provided.

1 - 18 . (canceled) 
     
     
         19 . A method performed by a computer comprising a processor for selecting genes to derive a gene-expression-based classifier, the method comprising:
 a) pooling a plurality of gene-expression data sets, wherein each of the plurality of data sets comprises a healthy control;   b) co-normalizing the gene-expression data of the pooled data set to remove batch effects by a correction factor derived from the healthy control of each dataset, wherein the correction factor removes technical variation across the plurality of data sets;   c) using a machine learning tool to identify at least two clusters within the co-normalized data;   d) assigning genes to the clusters, wherein a gene is assigned to a given cluster based on differential expression in the cluster compared to the other clusters or the healthy controls;   e) assessing the biological characteristics of each cluster based on gene ontology classification of the genes assigned to the cluster; and   f) iteratively incorporating the differentially expressed genes into a classifier for one or more of the clusters, wherein the differentially expressed genes are accepted into the classifier if the addition of the gene meets a threshold criteria for classifier performance, thereby selecting the genes to derive the gene-expression based classifier.   
     
     
         20 . The method of  claim 19 , wherein the classifier is validated in data sets without healthy controls. 
     
     
         21 . The method of  claim 19 , wherein the healthy controls are removed from the data sets after the pooled data is co-normalized. 
     
     
         22 . The method of  claim 19 , wherein the gene-expression data sets comprise diseased samples. 
     
     
         23 . The method of  claim 19 , wherein the gene-expression data sets comprise patients with sepsis. 
     
     
         24 . The method of  claim 19 , wherein the gene-expression data sets are obtained from samples comprising whole blood, white blood cells, neutrophils, or buffy coat. 
     
     
         25 . The method of  claim 19 , wherein the gene-expression comprises RNA expression levels. 
     
     
         26 . The method of  claim 19 , wherein the biological characteristics of each cluster are used to inform on the selection of a therapy. 
     
     
         27 . A method of treating a patient, comprising:
 a) selecting a therapy for a patient based on biological characteristics of the patient, wherein the biological characteristics are determined by which cluster the patient is assigned to using a classifier developed according to the method of  claim 19 ; and   b) administering the therapy.   
     
     
         28 . A method performed by a computer comprising a processor for training a gene-expression-based classifier, the method comprising:
 a) pooling a plurality of gene-expression data sets, wherein each of the plurality of data sets comprises a healthy control;   b) co-normalizing the gene-expression data of the pooled data sets to remove batch effects by a correction factor derived from the healthy control of each dataset, wherein the correction factor removes technical variation across the data sets;   c) identifying a plurality of subgroups among the patients in the pooled data sets;   d) selecting at least one gene of a classifier for assigning a patient to one of the subgroups based on differential expression of the at least one gene among the plurality of subgroups; and   e) training the classifier for subgroup assignment in the pooled co-normalized gene-expression data using cross-validation among the pooled co-normalized data.   
     
     
         29 . The method of  claim 28 , wherein the classifier is validated in datasets external to the plurality of pooled co-normalized datasets. 
     
     
         30 . The method of  claim 29 , wherein the validation comprises supervised classification. 
     
     
         31 . The method of  claim 19 , wherein the machine learning tool comprises an unsupervised analysis. 
     
     
         32 . The method of  claim 27 , wherein the therapy comprises administration of an innate or adaptive immunity modulator. 
     
     
         33 . The method of  claim 27 , wherein the therapy comprises administration of one or more drugs that modify the coagulation cascade or platelet activation. 
     
     
         34 . The method of  claim 27 , wherein the therapy comprises administration of a blood product, heparin, low-molecular-weight heparin, apixaban, dabigatran, rivaroxaban, dalteparin, fondaparinux, warfarin, activated protein C, recombinant coagulation cascade proteins, tranexamic acid, or another coagulation-modifying drug.