Development of a Machine Learning Prediction Model for the Detection of Meniere's Disease from Cerumen Chemical Profiles

阿尔茨海默病和相关痴呆补充申请的摘要/项目总结 [NOT-AG-22-025]（父项目：R21 DC 02056501） 阿尔茨海默病及其相关痴呆症（ADRD）困扰着全世界约5000万人。虽然测试 诊断和区分不同形式的痴呆症的方法已经研究了几十年，唯一的 确诊ADRD的确切方法是尸检。诊断是缓慢的，往往是基于排除标准 沿着而来的是多种形式的昂贵测试的结果。然而，如果更容易获得ADRD的化学标志物， 可以识别，快速和准确的诊断可以根据评估的存在（或不存在）， 相关化合物。这一成就将在方法和成本方面彻底改变ADRD诊断， 甚至揭示了疾病发病机制和进展的其他方面，可能揭示疾病的病因，并导致 更有效的替代疗法 这里假设，根据研究结果，ADRD部分表现为脂质的变化， 富含脂质的耳垢基质的化学特征可以作为ADRD存在的报告者，并且 耳垢特征可能因痴呆类型而不同。可以利用这些差异分布的知识， 通过将机器学习算法应用于化学品， 数据将通过追求以下具体目标对这一假设进行调查： 具体目标I：确定健康供体的耳垢的质谱衍生化学特征， 阿尔茨海默病（AD）患者和诊断为其他痴呆的患者。 具体目标二：开发机器学习预测模型，能够准确确定 阿尔茨海默病和/或其他痴呆，基于所有类型痴呆的共同特征，但与耳垢不同 健康的捐赠者 具体目标III：开发机器学习预测模型，以区分阿尔茨海默病样本和 其他类型的痴呆症使用耳垢化学概况，并确定的化合物的子集，是独特的， 每种类型的痴呆症 具体目标IV：开发的机器学习预测模型揭示的生物标志物的结构表征 具体目标二和三。 这项工作的结果将揭示是否有一个相关性之间的脂质配置文件的耳垢和存在， 阿尔茨海默氏病及相关痴呆症。我们将获得与之相关的分子的结构信息 健康人和痴呆症患者之间的差异。所披露的信息将为未来提供机会 开发一种潜在的非侵入性方法，用于快速诊断阿尔茨海默病和相关痴呆症。

ABSTRACT/PROJECT SUMMARY for Supplement Request on Alzheimer’s Disease and Related Dementias [NOT-AG-22-025] (Parent Project: R21DC02056501) Alzheimer’s disease and its related dementias (ADRD) afflict ~50 million people worldwide. Although testing methodologies to diagnose and differentiate different forms of dementia have been investigated for decades, the only definitive means to confirm a diagnosis of ADRD is at autopsy. Diagnosis is slow and is often based on exclusionary criteria along with the results of multiple forms of costly testing. However, if more readily accessible chemical markers of ADRD can be identified, rapid and accurate diagnoses could be accomplished based on assessment of the presence (or absence) of relevant compounds. Such an achievement would revolutionize ADRD diagnosis in terms of methods and cost, and could even reveal other dimensions of disease pathogenesis and progression that might shed light on disease etiology, and lead to alternative, more effective treatments. It is hypothesized here that based on research findings that reveal that ADRD manifests in part in terms of changes in lipid profiles, the chemical profile of the lipid-rich cerumen matrix may serve as a reporter of the presence of ADRD, and that cerumen profiles may differ as a function of dementia type. Knowledge of these differential profiles can be leveraged to accurately and rapidly reveal the presence of ADRD via the application of machine learning algorithms to the chemical data. This hypothesis will be investigated through pursuit of the following specific aims: Specific Aim I: Determination of the mass spectrum-derived chemical signatures of cerumen from healthy donors, Alzheimer’s disease (AD) patients, and patients diagnosed with other dementias. Specific Aim II: Development of machine learning prediction models that enable accurate determination of the presence of Alzheimer’s disease and/or other dementias based on features common to all types of dementia but distinct from cerumen from healthy donors. Specific Aim III: Development of machine learning prediction models to distinguish Alzheimer’s disease samples from other types of dementia using cerumen chemical profiles, and determination of the subset of compounds that are unique to each type of dementia. Specific Aim IV: Structural characterization of biomarkers revealed by the machine learning prediction model(s) developed in Specific Aims II and III. The results of this work will reveal whether there is a correlation between the lipid profile of cerumen and the presence of Alzheimer’s disease and related dementias. Structural information will be acquired on the molecules that are responsible for the differences in healthy and dementia patients. The information revealed would provide the opportunity for future development of a potential non-invasive method for the rapid diagnosis of Alzheimer’s disease and related dementias.