Identification of IGF-1-related genes in MI
By comparing the differential expression of myocardial infarction (MI) and the normal group (control). The analysis of differential expression identified 437 DEGs1 in the dataset GSE66360. Among these, 328 genes were found to be up-regulated, while 109 genes were down-regulated (Fig. 2A and B).
In the MI group of dataset GSE66360, MI patients were divided into high and low expression groups for differential analysis based on the median expression level of IGF-1, and a total of 1,592 DEGs2 were obtained, of which 762 were up-regulated and 830 were down-regulated (Fig. 2C and D).
Characterization of MI-related modules
All samples from GSE66360 datasets were clustered with no outliers detected (Fig. 3A). Afterwards, for the purpose to guarantee biologically meaningful scale-free topology, 12 was chosen as the minimal β values based on scale independence (R2 > 0.85). (Fig. 3B and C). Seven co-expression modules were identified (Fig. 3D). Of such, the green module (R = 0.61, p = 2e-11) had the highest correlation with MI, which comprising 479 genes (Fig. 3E).
Functional analysis and screening of candidate genes
Intersection of DEG1, DEG2 and module genes were taken by Wayne analysis to get 47 candidate genes were screened (Fig. 4A). Biological functions and signaling pathways of the candidate genes were analyzed by GO and KEGG enrichment. The 47 candidate genes were enriched for 255 biological functions (p < 0.05) (01.GO.csv),which related to biological processes such as neutrophil migration, myeloid leukocyte activation; The cellular components secretory granule membranes, autotertiary granules; and the molecular functions complement receptor activity (Fig. 4B). Enriched in 146 KEGG signaling pathways were candidate genes (p < 0.05), including the C-type lectin receptor signalling pathway, alcoholic liver disease, the IL-17 signalling pathway, and the TNF signalling pathway (Fig. 4C). In addition, PPI network findings included 47 node counts, 127 edge counts, and 15 outlier genes (Fig. 4D).
Identification to 5 hub genes and external validation
Using three Machine Learning algorithms, 5 hub genes (ALDH2, NLRP3, CLEC4E, ACSL1, and FCER1G) connected to MI were chosen among 47 candidate genes in this study.
Initialiy, the 17 feature genes were selected by LASSO analysis with the assurance of minimising the model error (λ = 0.02) (Fig. 5A).
Further, the SVM-RFE model was constructed in the training dataset to screen the hub genes. At the highest accuracy (Accuracy = 0.85), 36 feature genes were identified collectively (Fig. 5B). The feature genes screened by LASSO and SVM-RFE analysis were taken to intersect to obtain 14 candidate key genes (Fig. 5C).
RF classification model demonstrated that When ntree = 1000, the error within the model can be largely stabilized (Fig. 5D). The top 5 genes, ALDH2, NLRP3, CLEC4E, ACSL1, and FCER1G, were considered as hub genes in this study (Fig. 5E).
Finally, the hub genes in the training and validation datasets were validated, CLEC4E and ACSL1 were significantly upregulated (Fig. 5F and G).
Diagnostic performance of hub genes for MI
ROC curves and ANN diagnostic models were built to assess the diagnostic performance of the 5 hub genes for MI. The AUC of the 5 hub genes were all larger than 0.8 (Fig. 6A), indicating that they distinguished well between MI and healthy samples. The outcome of the ANN indicated that there was 1 hidden layer with 2 neurons (Fig. 6B). The confusion matrix for the ANN model was shown in Fig. 6C. Finally, the ROC curve was displayed to evaluate the the overall diagnostic ability of the 5 hub genes, and AUC = 0.903 revealed that the hub genes were stronger predictors of MI (Fig. 6D).
Impact of hub genes on the incidence of MI
Based on the expression of the 5 hub genes in the training set, an nomogram model of the hub genes was created (Fig. 6E). The ROC curve suggests that the nomogram model had good efficacy (AUC = 0.911) (Fig. 6F).
Signaling pathway analysis of hub genes
GSEA signaling pathways analysis revealed that the 5 hub genes enriched signaling pathways were mainly focused on the chemokine signaling pathway, Leishmania infection, Nod-like receptor signaling pathway and Toll-like receptor signaling pathway. These signaling pathways primarily participated in the modulation of immunological and inflammatory responses within the organism (Fig. 7A - E).
Analysis of ceRNA, TF-mRNA-miRNA and hub genes-related gene networks
The molecular regulatory network was visualized as shown in Fig. 7F and G. ACSL1 predicted a total of 7 miRNAs, 372 lncRNAs, and 1 TF (PPARA), whereas NLRP3 and CLEC4E each predicted 1 miRNA.
The GeneMANIA database analysis showed that 5 hub genes predicted 20 functionally related genes. The functions of functionally related genes focus on the regulation of innate immune responses, positive regulation of biostimulatory responses and regulation of IL-1β production (Fig. 7H).
Correlation of hub genes with MI and prediction of small-molecule drugs
Utilizing the CTD, the relationship between hub genes and MI was probed, and the results proved that the highest correlation score with ALDH2 was associated with MI, whereas the lowest was CLEC4E (Fig. 8A). Drug prediction revealed 4, 1 and 9 drugs interacting with 3 hub genes (ALDH2, FCER1G and FCER1G) respectively, while CLEC4E and ACSL1 did not predict the corresponding small-molecule drugs (Fig. 8B). The primary anticipated roles of small-molecule medicines are antibacterial and anti-inflammatory. For example, triclocarban targets NLRP3 and has an antibacterial effect, while anakinra and aspirin target FCER1G and NLRP3 respectively and have an inflammatory-reducing impact.
Hub genes significantly upregulated in blood samples from MI patients
The expression levels of hub genes were further validated in clinical samples and the results revealed that hub genes were significantly upregulated in MI, which provided powerful support for the results of this study (Fig. 9A - E).