Collapsing the list of myocardial infarction-related differentially expressed genes into a diagnostic signature

Abstract

Background: Myocardial infarction (MI) is one of the most severe manifestations of coronary artery disease (CAD) and the leading cause of death from non-infectious diseases worldwide. It is known that the central component of CAD pathogenesis is a chronic vascular inflammation. However, the mechanisms underlying the changes that occur in T, B and NK lymphocytes, monocytes and other immune cells during CAD and MI are still poorly understood. One of those pathogenic mechanisms might be the dysregulation of intracellular signaling pathways in the immune cells.

Methods: In the present study we performed a transcriptome profiling in peripheral blood mononuclear cells of MI patients and controls. The machine learning algorithm was then used to search for MI-associated signatures, that could reflect the dysregulation of intracellular signaling pathways.

Results: The genes ADAP2, KLRC1, MIR21, PDGFD and CD14 were identified as the most important signatures for the classification model with L1-norm penalty function. The classifier output quality was equal to 0.911 by Receiver Operating Characteristic metric on test data. These results were validated on two independent open GEO datasets. Identified MI-associated signatures can be further assisted in MI diagnosis and/or prognosis.

Conclusions: Thus, our study presents a pipeline for collapsing the list of differential expressed genes, identified by high-throughput techniques, in order to define disease-associated diagnostic signatures.

Keywords: Machine learning; Myocardial infarction; Transcriptional signatures; Transcriptomics; miRNA.

Fig. 1

A schematic pipeline of the study for MI transcriptional signatures’ identification. DEGs differentially expressed genes. MI patients with myocardial infarction. CTRLs individuals in the control group, CV cross-validation

Fig. 2

Volcano plot of gene expression changes in PBMC of MI patients compared to CTRLs. Blue dot indicates downregulated gene (log₂FC < −0.5); red dot indicates upregulated gene (log₂FC > 0.5), which passed threshold for multiple comparisons (p.adj < 0.05); Among differentially expressed genes (DEGs) MIR21 and its target genes are marked in orange, MIR223 and its target gene − in purple (−0.5 < log₂FC > 0.5, p < 0.05)

Fig. 3

Network analysis of the Reactome gene sets “Neutrophil degranulation” (a), “Cytokine Signaling in Immune system” (b) and “Immunoregulatory interactions between a Lymphoid and a non-Lymphoid cell” (c). The edges indicate molecular interactions between nodes based on String database

Fig. 4

Quality and robustness of the classification model with a L2-norm penalty function based on the cumulative expression levels of genes included in considered MI transcriptional signatures: {ADAP2}, {KLRB1 + KLRC1, KLRD1, KLRF1}, {MIR21 + BCL6, CCR1, PDGFD, TGFBR3, S100A12}, {MIR223 + MAFB} and {C3AR1, CD14, CR1, S100A12, SLC11A1}. a Areas Under receiver operating characteristic Curve (ROC-AUC) for the training (GSE59867) and test (GSE62646) datasets. b Time-depended (starting from MI onset) ROC-AUC metrics of the classification model

Fig. 5

Quality and robustness of the classification model with a L1-norm penalty function based on the cumulative expression levels of genes included in considered MI transcriptional signatures: {ADAP2}, {KLRB1 + KLRC1, KLRD1, KLRF1}, {MIR21 + BCL6, CCR1, PDGFD, TGFBR3, S100A12}, {MIR223 + MAFB} and {C3AR1, CD14, CR1, S100A12, SLC11A1}. a Coefficients of the classification model; the most important upregulated genes ADAP2, MIR21 and CD14 are marked in red, downregulated genes KLRC1 and PDGFD–in blue colour. b ROC-AUC metrics of the L1-regularized classification model consisted of ADAP2, MIR21 and CD14 genes. ROC-AUC were constructed using the training (GSE59867) and test (GSE62646) datasets. c Time-depended (starting from MI onset) ROC-AUC metrics of the L1-regularized classification model based on test dataset