Machine learning-based development of serologic test for acute Lyme disease diagnosis

Project Summary/Abstract Lyme Disease is a tickborne illness with markedly increasing prevalence in the United States and an urgent need for improved diagnostics in its early stages, when treatment is most efficient. While classic clinical presentation of the early illness is the presence of erythema migrans (EM), or “bullseye rash”, surrounding the tick bite site, 20-30% of patients do not present with EM. Further complicating diagnosis is a proportion of patients who present with EM, but are seronegative on the current standard two-tiered test algorithm (STTTA). The proposed research addresses the need for improved serological tests to diagnose early Lyme Disease in these patients, while the disease is the most responsive to treatment. A proof-of-concept antigen panel capable of distinguishing STTTA-positive acute Lyme samples from endemic controls was identified using a novel antigen discovery approach. This approach relies on representing an entire binding space of a donor’s circulating antibody repertoire using machine learning models based on the antibody binding profile to a diverse, random library of 126,050 peptides with an average length of 9 amino acids, which is a sparse representation of all possible amino acid combinations. Resulting models are then used to identify pathogen epitopes with high predictive power that are combined into a panel with diagnostic efficacy. Here, the unmet need of diagnosing early Lyme disease in STTTA-seronegative patients is addressed by the addition of antigens predicted as specific to this patient population. Diagnostic efficacy of the supplemented proof-of- concept antigen panel, that was identified in a previous proof-of-principle study, will be tested using an expanded cohort of STTTA seronegative donors and endemic controls. Specificity of the panel for Lyme disease will be confirmed using a panel of look-a-like illnesses including autoimmune diseases and tickborne diseases. This work is expected to yield data demonstrating the feasibility of a novel immunoassay for the diagnosis of early stage Lyme Disease patients currently missed by present tests. Additionally, it will serve as a demonstration of the antigen discovery approach as a means to identify diagnostic antigens for difficult pathogens.Project Narrative Improved diagnostics for early Lyme Disease, the most treatable stage of the disease, are urgently needed to identify infected patients, including those not presenting with the classic bullseye rash or who are seronegative on the standard two-tiered test algorithm. Here, we apply a novel antigen discovery approach that allows us to computationally model patients’ serum antibody binding profiles and predict diagnostic antigens for patients with early Lyme Disease. By developing a diagnostic assay which includes antigens specific for patients missed by current diagnostics, patient outcomes can be improved through enabling early intervention with effective treatments.