CAREER: Transforming Biosensor Reliability using Sensor Time-series Data and Physics-based Machine Learning

获得可靠的生物传感器可以通过帮助正在进行的和未来的大流行管理来改变公共卫生。然而，生物传感器的可靠性(例如假阳性(类型1)和假阴性(类型2)诊断)仍然是广泛应用于工业和临床的障碍。研究人员实验室进行的初步工作表明，使用生物传感器时间序列(TS)数据和基于物理的监督机器学习(ML)可以降低这些错误的概率。ML是一种根据数据进行预测的人工智能形式。因此，这个职业项目的研究目标是研究机器学习和化学工程领域知识的整合，以提高生物传感器的可靠性和性能。建议的方法将应用于各种传感器类型、大小、外形因素和数据结构。如果成功，获得可靠的生物传感器可以催化生物制造创新，并提高当前和新兴诊断方法的速度和准确性。该项目的教育目标是创建一个交互式开放课程软件(OCW)平台，为城市服务不足的学生增加医疗保健和数据科学领域的教育和劳动力发展机会。计划的活动包括为高中生提供以游戏驱动的生物传感模拟，为本科生举办关于传感器机器学习的数据存档的虚拟讲座和研讨会，为高中生和本科生举办关于机器学习在生物分析、生命和材料科学中的新兴应用的虚拟讲座。这位研究人员的首要职业目标是通过数据驱动的化学工程中的概念帮助改变生物传感器的性能，并扩大新兴数据驱动的生命科学行业中代表性不足的群体的领导力。为了与这一目标保持一致，本项目的目标是通过物理化学过程建模和监督ML的集成来改变生物传感器的可靠性。中心方法是将有监督的机器学习和传质受限的表面结合反应理论结合起来，通过生物传感器时间序列数据来提高生物分析物定量的可靠性。这个项目将测试这样一个假设，即将实验参数和传质限制表面结合反应理论与有监督的机器学习模型相结合，用于目标分析物分类，相对于最先进的校准方法，可以减少第一类和第二类错误的程度。建议的方法将被应用于可靠的基于生物传感器的RNA、microRNA和蛋白质靶标的检测，并以标准的临床生物分析方法为基准。这项工作将识别生物传感器时间序列数据中目标结合、非特异性结合和生物传感器漂移的新的数据和模型驱动的特征，这些特征可以支持使用机器学习对生物分析物浓度进行可靠的分类。如果成功，在生物传感器时间序列数据中识别目标结合和干扰输入的特征可以显著提高生物传感器和基于生物传感器的控制的可靠性和重复性。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Access to reliable biosensors could transform public health by aiding ongoing and future pandemic management. However, biosensor reliability (e.g. false positive (type 1) and false negative(type 2) diagnoses) remains a barrier to widespread industrial and clinical use. Preliminary work performed in the Investigator’s lab suggests that using biosensor time series (TS) data and physics-based supervised Machine Learning (ML), a form of artificial intelligence that makes predictions from data, can reduce the probability of these errors. Thus, the research goal of this CAREER project is to examine the integration of machine learning and chemical engineering domain knowledge for improving biosensor reliability and performance. The proposed methodology will be applied across various sensor types, sizes, form factors, and data structures. If successful, access to reliable biosensors could catalyze biomanufacturing innovations and improve the speed and accuracy of current and emerging diagnostic methods. The education goal of this project is to create an interactive Open Course Ware (OCW) platform to increase education and workforce development opportunities at the interface of healthcare and data sciences for urban-underserved students. Planned activities include Gaming-driven Simulations in Biosensing for High School Students, a Virtual Lecture and Workshop on Data Archiving for Sensor Machine Learning for Undergraduate Students and Virtual Lectures on Emerging Applications of Machine Learning in the Bioanalytical, Life, and Materials Sciences for High School and Undergraduate Students. The investigator’s overarching career goal is to help transform biosensor performance through concepts in data-driven chemical engineering and expand the leadership of underrepresented groups in emerging data-driven life sciences industries. In keeping with this goal, the objective of this project is to transform the reliability of biosensors through the integration of physiochemical process modeling and supervised ML. The central approach is to integrate supervised machine learning and mass transfer-limited surface binding reaction theory for improving the reliability of bioanalyte quantification via biosensor time-series data. This project will test the hypothesis that integrating experimental parameters and mass transfer-limited surface binding reaction theory with supervised machine learning models for target analyte classification can reduce the extent of type 1 and 2 errors relative to state-of-the-art calibration methods. The proposed methodology will be applied to reliable biosensor-based detection of RNA, microRNA, and protein targets and benchmarked against standard clinical bioanalytical methods. This work will identify new data- and model-driven features of target binding, nonspecific binding, and biosensor drift in biosensor time-series data that can support the reliable classification of bioanalyte concentration using machine learning. If successful, identifying features of target binding and interfering inputs in biosensor time-series data could significantly improve the reliability and reproducibility of biosensors and biosensor-based controls.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.