A-256 Multiplexed serologic test with advanced machine learning classifier for single-tier diagnosis of Lyme disease

Background Borrelia burgdorferi, the agent of Lyme disease, is estimated to cause nearly a half million diagnosed cases annually in the United States. The signs and symptoms of untreated Lyme disease have variable presentations that change as the disease progresses from early to later stages. Undiagnosed Lyme disease patients have a higher risk of developing a chronic illness which may be treatment resistant. Two-tier serology is the primary testing method currently used for Lyme disease diagnosis, but this approach has several limitations and shows poor clinical accuracy, particularly early in the disease. Alternative approaches are needed to improve the sensitivity of Lyme disease tests and enable correct diagnosis of Lyme disease patients. Methods We previously examined serially collected serum samples from non-human primates infected with B.burgdorferi and identified proteins that could be utilized for detection of infection at all phases of disease in humans. Additional protein targets were identified through epitope mapping for a wide range of B. burgdorferi antigens using peptide arrays. This approach facilitated the selection of a panel of serologic targets, which were then combined into a single fluorescent cytometric bead-based assay detecting both IgM and IgG antibodies using a Luminex 200 analyzer. 123 case and 197 control samples obtained from the SLICE study (Johns Hopkins University School of Medicine), the Lyme Disease Biobank (Bay Area Lyme Foundation), and the CDC were tested against this panel of antigens. Fluorescent values for all cases and controls were then interrogated with eight different advanced machine learning classifiers to identify which best distinguished between Lyme disease cases and controls. Results We identified a panel of 10 proteins and peptides representing primary antigens of B. burgdorferi that differentiated Lyme disease cases from controls. When combined into a single multiplexed immunoassay, interpretation of the combined IgM and IgG fluorescent values by the decision tree classifier XGBoost was found to show the highest overall performance. The receiver operating characteristic (ROC) analysis in Figure 1 shows the cross-validation performance of the assay with the XGBoost classifier. The classifier achieved an AUC of 0.978, compared to an AUC of 0.730 for the C6 antigen only, a conserved region of the B. burgdorferi variable surface protein VlsE which is the serological target used in the first tier of the current two-tier testing approach. Conclusion A multiplexed panel of antigens detecting both IgM and IgG antibodies in combination with the XGBoost classifier demonstrated improved test accuracy across all stages of Lyme disease. The complete assay significantly outperformed standard two-tier testing (STTT), and in the early Lyme/acute disease SLICE study cohort, a positive result was returned for 100% of early Lyme subjects, compared to only 27% positivity rate when tested by STTT. As a diagnostic test, this assay will provide clinically relevant information to healthcare providers to better inform treatment decisions or further diagnostic steps.