Construction Of A Diagnostic Model For Secondary Hemophagocytic Lymphohistiocytosis Based On Multi-omics Analysis And Machine Learning

Objective: hemophagocytic lymphohistiocytosis is a very dangerous disease with complex etiology,with rapid onset and high mortality,which poses a serious threat to the health of patients.However,the diagnosis of secondary hemophagocytic lymphohistiocytosis(sHLH)is still based on the diagnosis of primary hemophagocytic lymphohistio cytosis,and its pathogenesis is not fully understood.This study aimed to identify a predictive model for secondary hemophagocytic lympho histiocytosis by machine learning algorithm based on multi-omics.Methods: We collected whole blood from 8 sHLH patients and 8healthy people in our hospital.After centrifugation,monocytes were obtained for whole transcriptome sequencing.Serum was collected for proteomic and metabolomic analysis.The transcriptome data in the GEO/GTEx database was combined with the transcriptome data of our hospital,and the sequencing data of the sHLH group and the normal group were compared to obtain differential genes.Then,the pathways and functions involved in the core differential genes were investigated,and a ceRNA network was constructed.We also analyzed differential proteins and differential metabolites,and established three artificial neural network diagnostic models based on these differential genes,proteins,and metabolites.In addition,two machine learning algorithms were used to analyze the sequencing data to construct diagnostic models for methylation,ubiquitination,and acetylation,and use a nomogram to visualize the model for clinical use.Finally,we also performed immunoassays,as well as multi-omics conjoint analyses,to mine potential biomarkers.Results: By differential analysis we found that 3032 mRNAs(1193up-regulated and 1839 down-regulated),3491 circRNAs(1561 upregulated and 1930 down-regulated)and 5036 lncRNAs(2305up-regulated and 2731 down-regulated)were differentially expressed,and then A ceRNA network was constructed.The results of enrichment analysis showed that these differential genes were mainly involved in immune response,hematopoietic cell lineage,cell cycle,coagulation-related functions and pathways.A ten-gene diagnostic model(JTB,SNORD21,NADK,RNF220,KLHL17,RNU6 31 P,NBPF1,CDC73,RC3H1 and LAMTOR2)with ROC of 0.95 was constructed by artificial neural network algorithm.The diagnostic models of ten proteins and ten metabolites with ROC of 1 were also constructed.In addition,we used machine learning algorithms to build models from the perspective of methylation,ubiquitination,and acetylation,and constructed nomograms that are convenient for clinical use.All of them can be used for risk prediction of sHLH with good performance(AUC up to 1).Macrophage and eosinophilic infiltration were higher in immune cell assays.Through multi-omics analysis we found that LAMTOR2 may be a potential key biomarker.Conclusion: We constructed a 10-gene,10-protein,10-metabolite predictive model through multi-omics and machine learning algorithms to help clinicians diagnose sHLH.Enrichment analysis,immune correlation analysis and establishment of ceRNA network will help reveal the underlying molecular mechanism of sHLH.These results help us to better understand the disease characteristics and pathophysiological process of sHLH in many ways.