Discovery of early immunologic biomarkers for risk of PTLDS through machine learning-assisted broad temporal profiling of humoral immune response

项目总结/摘要 莱姆病（LD）是一种蜱传疾病，在美国的患病率明显增加。而更多 超过80%的LD病例可以使用已建立的抗生素治疗有效治疗，10-15%的患者 治疗后出现长期后遗症，表现为疲劳、认知障碍、关节疼痛等 症状，称为治疗后莱姆病综合征（PTLDS），导致患者经历 生活质量的实质性损失，并对患者和医疗保健造成显著的经济负担 系统由于缺乏经批准的分子诊断测试，许多医生认为， PTLDS完全，使患者的治疗选择很少或定义不清。虽然目前没有批准 PTLDS的治疗存在，新出现的证据表明，从早期开始， 在一些慢性和自身免疫性疾病中使用靶向治疗进行干预。早期风险评估 发展PTLDS提供了一个机会之窗，提醒患者和医生预测 一个长期的症状的结果和调整的基础上治疗。拟议研究的重点是 迫切需要鉴定用于早期预测患者发生PTLDS风险的免疫学生物标志物 在疾病的急性期，并有可能显着改善临床结果， 早期干预。所提出的方法从一个全面的 患者循环抗体库的结合谱。该方法的新奇之处在于， 供体循环抗体库的整个结合空间，而不是简单地关注先验已知的 抗原靶点。该方法依赖于使用在抗体结合上训练的机器学习模型 将这些肽的平均长度为9-10个氨基酸的126，050个独特肽的多样性随机文库作为一个序列， 所有可能的组合9聚体序列的稀疏表示。预测模型用于 鉴定具有高预测能力的疾病相关病原体表位， PTLDS风险评估小组。假设B.布尔格。抗原和/或自身抗体 涉及人类蛋白质组，并且有一组可用作生物标志物来预测 PTLDS在疾病早期。将在一组纵向收集的 患者样本，因为他们从确诊的急性LD进展，通过治疗疾病的结果。这 应用的方法将使抗体反应的广度，包括潜在的反应/交叉反应， 与待检查的人蛋白质的反应性沿着抗体应答的异质性， 患者队列。蛋白抗原性的计算机模拟预测将使用基于溶液的 免疫测定。拟议的工作预计将确定一套B。布尔格。和潜在的人类自身免疫 与PTLDS进展相关的抗原。这些知识将作为基础， 未来的诊断、治疗或产生可以在疾病模型中测试的假设。

Project Summary/Abstract Lyme Disease (LD) is a tickborne illness with markedly increasing prevalence in the United States. While more than 80% of the LD cases can be effectively treated using established antibiotic treatments, 10-15% of patients develop post-treatment, long-term sequalae manifested as fatigue, cognitive impairment, joint pain, and other symptoms and termed Post Treatment Lyme Disease Syndrome (PTLDS) causing patients to experience substantial loss in quality of life and resulting in a marked financial burden on both patient and health care system. The absence of approved molecular diagnostic tests leads many physicians to dismiss the notion of PTLDS entirely, leaving patients with few or poorly defined treatment options. While currently no approved treatment for PTLDS exists, emerging evidence suggests substantially better clinical outcomes from early intervention with targeted therapies in a number of chronic and autoimmune disorders. Early risk assessment of developing PTLDS offers a window of opportunity by alerting both the patient and the physician to anticipate a long-term symptomatic result and adjust symptom-based treatment. The proposed study focuses on the urgent need to identify immunologic biomarkers for predicting the risk of a patient developing PTLDS early during the acute phase of the disease and has the potential to markedly improve clinical outcomes through early intervention. The proposed approach derives disease-specific antigen information from a comprehensive binding profile of the patient’s circulating antibody repertoire. The novelty of the approach is in representing an entire binding space of a donor’s circulating antibody repertoire, instead of simply focusing on a priori known antigenic targets. The approach relies on using machine learning models trained on the antibody binding profile to a diverse, random library of 126,050 unique peptides with an average length of 9-10 amino acids as a sparse representation of all possible combinatorial 9-mer sequences. Predictive models are then used to identify disease-associated pathogen epitopes with high predictive power that can be combined into a potential panel for PTLDS risk assessment. It is hypothesized that B. burg. antigens and/or self-antibodies from the human proteome are involved and that there is a set that can be used as biomarkers to predict progression to PTLDS early in the disease. The antibody response over time will be profiled in a set of longitudinally collected patient samples as they progress from confirmed acute LD, through treatment to disease outcome. This approach applied will enable the breadth of the antibody response, including a potential response/cross- reactivity to human proteins to be examined along with the heterogeneity of antibody responses across a cohort of patients. In silico predictions of protein antigenicity will be confirmed using solution-based immunoassays. The proposed work is expected to identify a set of B. burg. and potentially human autoimmune antigens that are associated with progression to PTLDS. Such knowledge is expected to serve as the basis for future diagnostics, therapeutics or in the generation of hypothesis that can be tested in disease models.