The application provides a serum
proteomics and
machine learning-based
Kawasaki disease intravenous immunoglobulin (IVIG) resistance biomarker combination and
screening method, and belongs to the technical field of medicines. The method comprises the following steps: (1) on the basis of establishing strict inclusion criteria and
typing criteria, collecting
whole blood samples of IVIG
reaction type and non-
reaction type Kawasaki disease children before treatment; (2) using DIA
proteomics technology for systematic screening and differential
protein identification; (3) weighted co-expression
network analysis, screening of
protein modules significantly related to IVIG non-reaction
phenotype; (4) combined with LASSO-
Logistic regression and SVM-RFE for multi-step
feature screening, identifying five biological markers significantly related to IVIG resistance: PLA2G4A, SNX17, PURB, CERS3 and CASP1 (5) based on the marker expression level, using the pROC
package for ROC analysis and calculating AUC; (6) analyzing the correlation between the marker and the clinical index related to
Kawasaki disease; (7) after limma
processing in the independent
transcriptome set GSE18606, using glm to construct a multivariate binary
logistic regression prediction model and perform ROC analysis. Through independent
transcriptome dataset
verification, the biomarker combination screened by the application can realize effective prediction of Kawasaki
disease IVIG resistance, and shows good prediction performance and clinical application value.