Diagnostic model based on bioinformatics and machine learning to distinguish Kawasaki disease using multiple datasets

BACKGROUND: Kawasaki disease (KD), characterized by systemic vasculitis, is the leading cause of acquired heart disease in children. Herein, we developed a diagnostic model, with some prognosis ability, to help distinguish children with KD. METHODS: Gene expression datasets were downloaded from Gene Expression Omnibus (GEO), and gene sets with a potential pathogenic mechanism in KD were identified using differential expressed gene (DEG) screening, pathway enrichment analysis, random forest (RF) screening, and artificial neural network (ANN) construction. RESULTS: We extracted 2,017 DEGs (1,130 with upregulated and 887 with downregulated expression) from GEO. The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses showed that the DEGs were significantly enriched in innate/adaptive immune response-related processes. Subsequently, the results of weighted gene co-expression network analysis and DEG screening were combined and, using RF and ANN, a model with eight genes (VPS9D1, CACNA1E, SH3GLB1, RAB32, ADM, GYG1, PGS1, and HIST2H2AC) was constructed. Classification results of the new model for KD diagnosis showed excellent performance for different datasets, including those of patients with KD, convalescents, and healthy individuals, with area under the curve values of 1, 0.945, and 0.95, respectively. CONCLUSIONS: We used machine learning methods to construct and validate a diagnostic model using multiple bioinformatic datasets, and identified molecules expected to serve as new biomarkers for or therapeutic targets in KD. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12887-022-03557-y.