2026-10-11T18:19:37Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34029708 2026-09-30T04:47:52Z category_696 portal_316 item_type_3 month_year_09_2026

Table 1_A lipid-related transcriptomic signature for distinguishing Kawasaki disease from febrile infections: a Mendelian randomization and machine-learning study.xlsx Sibao Wang (216314) Gang Luo (425896) Zhixian Ji (15270463) Guoxiang Zhou (17411769) Silin Pan (15201991) Genetic Immunology differential diagnosis endothelial activation Kawasaki disease lipid metabolism machine learning Mendelian randomization Background<p>Kawasaki disease (KD) diagnosis remains challenging because no pathognomonic test exists and the clinical presentation overlaps extensively with common childhood infections. Acute KD is accompanied by reproducible lipid perturbations, yet whether these changes are causally linked to the innate immune-driven vascular damage of KD vasculitis is unestablished. We aimed to derive a lipid-related transcriptomic signature for distinguishing KD from febrile infections and to characterize its relationship with innate immune remodeling.</p>Methods<p>Bidirectional two-sample Mendelian randomization (MR) screened 422 candidate exposures (419 lipid/metabolic traits and three immune-cell phenotypes) against two independent KD GWAS cohorts, with multiplicity correction. Differentially expressed genes from a training cohort comparing KD with rigorously classified febrile controls were intersected with a broad lipid-metabolism-related GeneCards set, yielding 459 candidates. The 459 candidates were evaluated across 130 machine-learning combinations; a full-training LASSO fit at lambda.1se yielded a 12-gene candidate panel, and the saved penalized model was applied unchanged to an independent cohort without batch correction. The immune microenvironment was estimated using CIBERSORT, acute-to-convalescent expression changes were tracked longitudinally, and selected markers were assessed in TNF-α-stimulated HUVECs.</p>Results<p>No exposure survived correction for the 422 tests performed (minimum Benjamini–Hochberg q = 0.979). All nine discovery-stage nominal hits were evaluated in a second KD GWAS; eight had directionally concordant point estimates, but none reached an unadjusted P < 0.05 in the second cohort. The frozen penalized model achieved an external AUC of 0.851 (95% CI 0.768–0.934) in GSE68004 without cross-cohort batch correction. CIBERSORT indicated a higher estimated neutrophil fraction in KD, concordant with a mature-neutrophil marker score from the same matrix. CETP, KREMEN1, and CCL23 were lower in convalescent samples after IVIG (all FDR < 0.0001) and were induced in TNF-α-treated HUVECs, together with VCAM-1, ICAM-1, and IL-8.</p>Conclusions<p>The frozen penalized model showed moderate discrimination between KD and febrile controls, but the selected 12-gene set represents one candidate realization of a partition-sensitive modeling pipeline. KREMEN1 warrants separate biological study on the basis of marginal, longitudinal, and immune-association findings rather than its selection stability. Recalibration and prospective validation are required before clinical use. The HUVEC findings are transcriptional and hypothesis-generating.</p> 2026-09-30T04:47:52Z Dataset Dataset 10.3389/fimmu.2026.1851332.s002 https://figshare.com/articles/dataset/Table_1_A_lipid-related_transcriptomic_signature_for_distinguishing_Kawasaki_disease_from_febrile_infections_a_Mendelian_randomization_and_machine-learning_study_xlsx/34029708 CC BY 4.0