多発性硬化症の神経ストレス免疫シグネチャーのマルチデータセットからの機械学習モデリングによる早期予測

Alopecia areata (AA) is an easy-recurring disease that presents huge challenges globally. An efficient clinical tool to predict AA onset would be valuable for timely intervention. We extracted six AA-related datasets from Gene Expression Omnibus (GEO). GO, KEGG, GSEA, GSVA and CIBERSORT algorithm were performed to elucidate the characteristics of AA. Feature genes were identified using LASSO regression and Random Forest algorithms. Five machine learning algorithms (Logistic Regression, K-nearest neighbors, Elastic Net, XGBoost and LightGBM) were employed to construct predictive models, with internal and external validation conducted to determine the optimal model. Additionally, SHapley Additive exPlanations (SHAP) analysis was applied to interpret the best-performing model and shiny framework was applied to establish an online predictive website. Five datasets (GSE45512, GSE68801, GSE80342, GSE58573, GSE74761) were integrated as train set and GSE148346 was defined as test set. Tissue regeneration and immune dysregulation were the key factors in AA pathogenesis. Three feature genes (KRT83, PPP1R1C, PIRT) were selected for model construction, with innate immune response, neural inflammatory and stress being a potential regulator for AA. The XGBoost model outperformed other algorithms, SHAP provided explanations for predictions and an online predictive website was established. Our study provides a potential "neuro-stress-immune" interplay insight into the pathogenesis of AA and establishes a clinically applicable predictive model for AA onset.