Collaborative Research: Improving the Performance and Design of Potentiometric Biosensors for the Detection of Extracellular Histones in Blood with Deep Learning

护理点(POC)传感器旨在随时随地为患者和医生提供所需的诊断信息。电位型生物传感器输出的电压是目标生物分子浓度的函数，非常适合对灵敏度、检测速度、便携性和与低功率读出电路的兼容性都非常重要的POC使用。不幸的是，这种传感器的大多数演示在从测试受控良好的实验室解决方案到在血清或全血中运行的转换过程中停滞不前。这项研究的目标是克服这一障碍，方法是使用一类名为深度学习的高级人工智能技术来识别从基于血液的测试收集的复杂数据中的模式和关系，以提高灵敏度并推动这些传感器的设计优化。这种方法将被应用于血液中循环组蛋白的检测，这些组蛋白有助于危重患者发生多器官功能障碍综合征(MODS)。据估计，在美国所有重症监护病房(ICU)入院的患者中，有15%会导致MODS，这给医疗保健系统造成了数十亿美元的损失。目前，还没有生物标记物来识别那些MODS风险增加的患者。拟议的深度学习增强型组蛋白传感器的成功开发将使患者能够及早识别，这些患者将受益于更积极和有针对性的治疗，以预防MODS和相关并发症。这些概念将与高中生的可穿戴设备挑战相结合，数据也将包括在本科和研究生课程中。拟议的研究包括回答以下科学和工程问题：(1)传统的RNA适配子功能化电位器对循环组蛋白的检测极限和反应速度是什么？针对组蛋白的RNA适配子将被用于金敏电极的功能化，以首次建立能够早期识别MODS的扩展门电位器的检测极限和响应速度。这些设备将在缓冲液、血清和全血中进行评估，作为部署POC的基准。(2)深度学习如何提高电位型生物传感器在评估全血时的性能？电位型生物传感器的性能依赖于几个因素(如电极选择、表面功能化、样品类型等)，这使得将其转化为血液分析是一个重大挑战。我们将利用深度学习技术来揭示复杂的关系和趋势，以弥补在基于血液的测试中观察到的传统灵敏度损失。这些发现还将推动电位传感器的优化设计，从而建立设计规则，加快这些传感器在整个社区的发展。(3)为深度学习开发训练数据的最佳方法是什么？将机器/深度学习技术应用于生物传感的一个主要障碍是生成足够的训练数据。将开发一个多路电位生物传感平台，通过使用扩展GATE方法使之成为可能，以便找出节省时间和资源的算法培训办法。这一努力将建立一个标准化的协议，该领域的其他研究人员可以利用它来加速电位生物传感器在新应用中的采用。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。

Point of care (POC) sensors aim to provide patients and medical practitioners with diagnostic information when and where it is needed. Potentiometric biosensors, which output a voltage as a function of target biomolecule concentration, are ideally suited for POC use in which sensitivity, speed of detection, portability and compatibility with low-power read-out circuitry are all paramount. Unfortunately, most demonstrations of such sensors stall during the translation from testing well-controlled laboratory solutions to operating in serum or whole blood. The objective of this research is to overcome this hurdle by using a class of advanced artificial intelligence techniques known as Deep Learning to recognize patterns and relationships in the complex data that is collected from blood-based tests to improve sensitivity and drive the design optimization of these sensors. This approach will be applied to the detection of circulating histones in blood, which contribute to the development of Multiple Organ Dysfunction Syndrome (MODS) in critically ill patients. It is estimated that 15% of all intensive care unit (ICU) admissions in the United States result in MODS, costing the healthcare system billions of dollars. Currently, there is no biomarker to identify those patients at increased risk of MODS. The successful development of the proposed Deep Learning-enhanced histone sensor will allow for the early identification of patients that will benefit from more aggressive and targeted therapies to prevent MODS and related complications. These concepts will be integrated with wearable device challenges for high school students, and data will also be included in undergraduate and graduate curricula.The proposed research consists of answering the following scientific and engineering questions: (1) What is the conventional limit of detection and speed of response of RNA aptamer-functionalized potentiometers to circulating histones? RNA aptamers specific to histones will be used to functionalize gold sensing electrodes to establish, for the first time, the limit of detection and speed of response of extended gate potentiometers capable of early identification of MODS. These devices will be evaluated in buffer, serum and whole blood as benchmarks for POC deployment. (2) How can Deep Learning improve the performance of potentiometric biosensors beyond their conventional limits when assessing whole blood? Potentiometric biosensor performance relies on several factors (e.g., electrode choice, surface functionalization, sample type, etc.), which make their translation to blood analysis a major challenge. We will leverage deep learning techniques to reveal intricate relationships and trends to compensate for the conventional losses in sensitivity observed in blood-based tests. These findings will also drive the optimal design of the potentiometric sensors, thus establishing design rules that can accelerate the development of these sensors across the community. (3) What is the optimal method to develop training data for deep learning? A major obstacle to the application of Machine/Deep Learning techniques to biosensing is the generation of adequate training data. A multiplexed potentiometric biosensing platform, made possible by the use of the extended gate approach, will be developed in order to identify time- and resource-efficient approaches to algorithm training. This effort will establish a standardized protocol that other researchers in the field can leverage in order to accelerate the adoption of potentiometric biosensors in new applications.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.