91 - AI-DRIVEN DISCOVERY OF TWO-GENE BLOOD TEST FOR THE DIAGNOSIS OF KAWASAKI DISEASE USING A CAP ACCREDITED CLIA CERTIFIED QPCR PLATFORM

Background

Kawasaki disease (KD) poses a significant diagnostic challenge due to its resemblance to other febrile illnesses and its potential to cause serious heart issues if untreated. Currently, there is no clinically available objective molecular diagnostic test. To address this, we used AI-assisted analysis of large public datasets to identify KD diagnostic biomarkers. The biomarkers were further validated using qPCR in four independent KD cohorts.

METHODS

In-Silico Analysis: Differentially expressed genes (DEGs) were identified and categorized using biological pathway analysis and refined using human-biofluid proteome databases. Additionally, twelve genes were sourced from literature reports on gene expression variances in KD versus febrile controls (FCs). DEGs gene pair ratio and AI-assisted analysis were further performed on independent datasets to identify the best gene pair to discriminate KD from febrile controls. Quantitative PCR (qPCR) validation: The DEG biomarker gene pair was subject to qPCR validation analysis using retrospective KD and febrile control samples. The gene pair's Delta C(t) was calculated and used as KD risk scores to validate the KD biomarkers.

RESULTS

AI-assisted in silico analysis finalized a DEG pair capable of discriminating KD from febrile control with an ROC AUC of 0.96; Additionally, four KD cohorts were used to validate the gene pair using qPCR assays. The initial discovery cohort yielded an AUC of 0.946. Using a two-score cutoff system, the two-gene scoring can identify high-risk KD with a PPV of 94.3% and low-risk FC patients with an NPV of 97.0%. This gene pair was further validated using a separate cohort of 60 KD and 60 FC samples, achieving an AUC of 0.910. The assay accurately identified KD with high-risk PPV at 89.3% and low-risk FC NPV at 97.3% using cutoffs from the initial discovery cohort. The KD assay also identified all patients with coronary artery aneurysms (Z-score > 2.5) and all dilated coronary artery patients. Two additional independent cohorts were also tested with a PPV of 90.9% and an NPV of 100% in the CAP/CLIA cohort and a PPV of 83.3% and NPV of 80.0% in a whole blood cohort.

CONCLUSION

This study developed a simple two-gene expression panel using a common qPCR platform. Through our CAP/CLIA validation process, we demonstrate that hospitals could quickly adapt this assay for direct testing in whole blood. The assay can also be easily converted to a near-patient or point-of-care device based on qPCR technology.