Predictive modeling of ischemic stroke, identification of relevant subgroups, and drug prediction based on multiple programmed cell death patterns

Abstract

Abstract Background: Ischemic stroke is the second leading cause of mortality and disability globally. Along with many immune and disease conditions, Programmed cell death (PCD) also has a critical role in ischemic stroke and may serve as a diagnostic indicator of ischemic stroke. Methods: From the Gene Expression Omnibus database (GEO), two ischemic stroke datasets were chosen, one for training and the other for the validation group. From the KEGG and other databases, 12 patterns of PCD-related genes were selected. Differentially expressed genes (DEG) were found using Limma analysis; functional enrichment analysis;machine learning least absolute shrinkage and selection operator (LASSO) regression; candidate immune-related central genes were identified using Random Forest along with the construction of a protein-protein interaction network (PPI) and an artificial neural network (ANN) for validation. In order to diagnose an ischemic stroke, the Receiver operating characteristic (ROC) curve was plotted, the diagnostic model was validated by qRT-PCR, immune cell infiltration was investigated to observe immune cell dysregulation in ischemic stroke, and the expression of candidate models under different isoforms was analyzed by consensus clustering (CC). Finally, drugs associated with candidate genes were collected through the Networkanalyst online platform. Results: A total of 71 genes were shown to be the crossover of DEG and PCD-related genes in ischemic stroke, and six candidate genes were finally identified by machine learning to establish a diagnostic prediction model. After using an artificial neural network (ANN) for validation, ROC curve plotting, and qRT-PCR validation for diagnostic value assessment. The outcomes demonstrated that the prediction model had a high diagnostic value. In the immune infiltration analysis, significant variability of NKT was found in ischemic stroke patients. Seven drugs associated with candidate genes were collected from the Networkanalyst online platform. Conclusion: A diagnostic prediction model with a good effect in the training group and validation group (AUC 0.94, CI 1.00-0.88 and AUC 0.91, CI 0.97-0.86, respectively), along with a good phenotype in qRT-PCR validation by comprehensive analysis was obtained. Additionally, the drugs (C646 substance, Cyclosporine, Decitabine, Dexamethasone, Resveratrol, Silicon Dioxide, and Tretinoin) that might be useful in the treatment of ischemic stroke were obtained.