RESUMO
MIS-C is a severe hyperinflammatory condition with involvement of multiple organs that occurs in children who had COVID-19 infection. Accurate diagnostic tests are needed to guide management and appropriate treatment and to inform clinical trials of experimental drugs and vaccines, yet the diagnosis of MIS-C is highly challenging due to overlapping clinical features with other acute syndromes in hospitalized patients. Here we developed a gene expression-based classifier for MIS-C by RNA-Seq transcriptome profiling and machine learning based analyses of 195 whole blood RNA and 76 plasma cell-free RNA samples from 191 subjects, including 95 MIS-C patients, 66 COVID-19 infected patients with moderately severe to severe disease, and 30 uninfected controls. We divided the group into a training set (70%) and test set (30%). After selection of the top 300 differentially expressed genes in the training set, we simultaneously trained 13 classification models to distinguish patients with MIS-C and COVID-19 from controls using five-fold cross-validation and grid search hyperparameter tuning. The final optimal classifier models had 100% diagnostic accuracy for MIS-C (versus non-MIS-C) and 85% accuracy for severe COVID-19 (versus mild/asymptomatic COVID-19). Orthogonal validation of a random subset of 11 genes from the final models using quantitative RT-PCR confirmed the differential expression and ability to discriminate MIS-C and COVID-19 from controls. These results underscore the utility of a gene expression classifier for diagnosis of MIS-C and severe COVID-19 as specific and objective biomarkers for these conditions.
