Surface plasmon resonance imaging (SPRi) in combination with machine learning for microarray analysis of multiple sclerosis biomarkers in whole serum
Surface plasmon resonance imaging (SPRi) in combination with machine learning for microarray analysis of multiple sclerosis biomarkers in whole serum
复制标题
DOI：
10.1016/j.biosx.2022.100127
复制
发表时间：
2022-05
影响因子：
--
通讯作者：
Alexander S. Malinick;Daniel D. Stuart;Alexander S. Lambert;Q. Cheng
中科院分区：
文献类型：
--
作者：
Alexander S. Malinick;Daniel D. Stuart;Alexander S. Lambert;Q. Cheng
多发性硬化症（MS）是在年轻人中观察到的最常见的自身免疫性疾病，并且已知非常难以准确诊断。目前的诊断方法被认为是不可靠和低效的，并且它们通常缺乏允许常规监测疾病进展所需的特异性。在这项工作中，我们报告了一种表面等离子体共振成像（SPRi）方法与碳水化合物微阵列相结合，用于检测未稀释血清中的多发性硬化症生物标志物。证明工作范围为1-100 ng/mL，检测限（LOD）低于7 ng/mL。在这项工作中使用的微阵列被涂覆有全氟癸基三氯硅烷（PF 4S）与神经节苷脂抗原的疏水性尾部强烈相互作用，允许以模仿髓鞘的方式进行理想的抗原展示。将机器学习（ML）算法应用于碳水化合物阵列/SPRi数据分析，以理解和表征在抗体之间观察到的交叉反应性。使用统计模型分析终点结果和SPRi传感图，以评价包括动力学和稳态组分的结合事件。此外，K-最近邻（kNN）和神经网络（nnet）被用来检查特异性和交叉反应性结合，产生更高的准确性比传统的方法可以实现。ML模型和微阵列数据的结合提供了对复杂相互作用的全面理解，并可用于在临床环境中区分和识别行为密切的生物标志物。
Multiple sclerosis (MS) is the most common autoimmune disease observed in young adults and is known to be exceptionally difficult to diagnose accurately. Current diagnostic methods are considered unreliable and inefficient, and they typically lack the needed specificity that allows for routine monitoring of disease progression. In this work, we report a surface plasmon resonance imaging (SPRi) method in combination with carbohydrate microarrays for the detection of multiple sclerosis biomarkers in undiluted serum. A working range of 1–100 ng/mL was demonstrated with the limit of detection (LODs) below 7 ng/mL. The microarrays utilized in this work were coated with perfluorodecyltrichlorosilane (PFDTS) to interact strongly with the hydrophobic tails of the ganglioside antigens, allowing for desirable antigenic display in a manner mimicking a myelin sheath. Machine learning (ML) algorithms were applied to the carbohydrate array/SPRi data analysis to understand and characterize the cross reactivities observed between the antibodies. Both endpoint results and SPRi sensorgrams were analyzed with statistical models for the evaluation of binding events that include kinetic and steady state components. In addition, K-nearest neighbor (kNN) and neural net (nnet) were utilized to examine specific and cross-reactive binding, yielding higher accuracy than what traditional methods can achieve. The combination of ML models and microarray data provides a comprehensive understanding of complex interactions and could be used to differentiate and identify closely behaving biomarkers in a clinical setting.