Machine learning screening for molecular subtyping of Alzheimer's disease based on pyroptosis-related genes

Abstract

Abstract The aim of the present research was to find the Alzheimer's disease (AD) disease characteristic genes (DCGs) by machine learning algorithms, and constructed and verified the nomogram model of AD based on DCGs. In this study, Gene Expression Omnibus (GEO) database GSE132903 dataset was initially downloaded and sorted out a gene expression matrix for AD. Subsequently, through differential analysis of the gene expression matrix, We discovered 34 DE-PRGs, or differentially expressed pyroptosis-related genes. The expression levels of DE-PRGs in people with Alzheimer's were correlated using correlation analysis, we have gained insights into the interconnected regulatory dynamics among these DE-PRGs. On top of that, we classified the 97 AD samples into two molecular subtypes using cluster analysis based on the expression patterns of 34 DE-PRGs. We created the extreme gradient boosting (XGB), support vector machine (SVM), random forest (RF), and generalised linear model (GLM) models based on the expression features of these DE-PRGs to further select them as having high diagnostic value. In order to forecast illness risk, the SVM model was used and a nomogram was constructed to evaluate its predictive accuracy. Moreover, we validated the diagnostic performance of the model for predicting AD by calibrating the curve, decision curve analysis (DCA) curve, and independent validation dataset GSE5281. Together, our findings may provide new insights into risk prediction, early diagnosis, and targeted therapy for AD in the population.