Deep learning-based serological test for point-of-care analysis of COVID-19 immunity with a paper-based multiplexed sensor

基于深度学习的血清学测试，使用纸基多重传感器对 COVID-19 免疫进行即时分析摘要：由 SARS-CoV-2 病毒引起的 COVID-19 于 2020 年 3 月 12 日被世界卫生组织 (WHO) 宣布为大流行病。诊断测试一直是应对措施的关键焦点，迫切需要快速开发、扩展和分发新的测试。尽管为直接检测 SARS-CoV-2 遗传物质开发了所有成功的测试方法，但仍然迫切需要创建能够检测病毒特异性抗体的新血清学测定方法，因为它们可以通过指示先前的暴露和潜在的免疫力来确定直接检测方法的补充信息，这由于各种新出现的变体而显得尤为重要。此外，随着针对新变种的疫苗推出，这些血清学测试可用于评估疫苗接种活动的效果，包括在已接种疫苗和未接种疫苗的人群中引发 SARS-CoV-2 和变种抗原特异性抗体的能力。与目前的直接检测方法相比，检测抗体的血清学测试成本较低，并且有利于即时护理 (POC) 设置，从而能够进行广泛的筛查工作，例如广泛的免疫测试，以表明个人需要疫苗加强剂，使个人有资格旅行、重返工作岗位和/或识别康复期血浆捐献者。为了满足这一迫切需求，该项目将创建一个基于智能手机的经济高效的平台，该平台可以感知和测量一个人可能产生的多种针对 SARS-CoV-2 的不同抗体，其测试格式易于使用，并且可以使用廉价的纸质测试在 15 分钟内完成。研究人员团队将开发一种多重 POC 免疫测定和血清诊断算法，该算法将从多达 10 种独特的免疫反应中推断疫苗接种/免疫状态，以区分一系列 SARS-CoV-2 抗体。为此，研究小组将创建一种多重垂直流测定（xVFA），以同时检测S蛋白的IgA、IgM和IgG抗体（以及S蛋白的变体，如δ、lambda和其他新出现的变体），并具有专门针对SARS-CoV-2病毒及其最新变体中S-蛋白的S-1、S-2和受体结合结构域（RBD）的独立免疫反应位点。研究团队将使用现有的和去识别化的人血清样本，通过xVFA平台筛查COVID-19阳性样本，包括常见变异产生的样本（通过逆转录酶-聚合酶链反应和测序确认）以及接种疫苗的样本和大流行前未接种疫苗的阴性对照样本。然后，将使用来自多重免疫反应的定量信息和一组剩余人类血清样本的真实临床状态来训练神经网络。此训练阶段将 (1) 创建血清诊断算法，以使用多重抗体测量来识别对 SARS-CoV-2 感染（包括常见变异）的阳性免疫反应或疫苗接种状态，以及 (2) 识别抗体-抗原相互作用的关键子集，以最准确地表示和量化对 SARS-CoV-2 感染或通过疫苗接种进行的保护的免疫反应。盲测试阶段将对多路复用和数据驱动方法的性能增强进行基准测试，以严格验证经过训练的推理网络的泛化能力。通过验证用于 COVID-19 免疫保护的新型多重垂直流测定和血清诊断算法，研究团队旨在确定通过多次测量和计算分析获得的灵敏度和特异性的显着改进，而几乎不需要增加成本或操作步骤或所需的样本量。该项目还将建立一个补充性的教育推广计划，其中包括（1）新闻媒体和互联网上的公开采访和科普文章； (2) 涉及代表性不足的学生的本科生研究机会； (3) 通过组织讲习班、研讨会和会议进行研究生培训。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Deep learning-based serological test for point-of-care analysis of COVID-19 immunity with a paper-based multiplexed sensorAbstract: COVID-19, caused by the virus SARS-CoV-2, was declared a pandemic by the World Health Organization (WHO) on March 12, 2020. Diagnostic testing has been a critical focus of the response, with an urgent need to rapidly develop, scale, and distribute new tests. Despite all the successful testing methods developed for the direct detection of SARS-CoV-2 genetic material, there is still an urgent need to create new serological assays that can detect virus-specific antibodies as they can ascertain complementary information to direct detection methods by indicating previous exposure and potential immunity, especially important due to various emerging variants. In addition, as vaccines against new variants roll out, these serological tests can be used to evaluate the efficacy of vaccination campaigns, including the ability to elicit SARS-CoV-2 and variant antigen-specific antibodies across vaccinated and unvaccinated populations. In contrast to the current direct detection methods, serology tests that detect antibodies can be low-cost and conducive to a point-of-care (POC) setting, enabling broad screening efforts like widespread immunity testing to indicate individuals in need of vaccine boosters, qualify individuals for travel, return to work, and/or identify convalescent plasma donors. To serve this urgent need, this project will create a smartphone-based, cost-effective platform that can sense and measure the many different antibodies specific to SARS-CoV-2 a person may develop, in a testing format that is easy to use and can be completed within 15 min using an inexpensive paper-based test. The team of researchers will develop a multiplexed POC immunoassay and serodiagnostic algorithm that will infer the vaccination/immunity status from up to 10 unique immunoreactions to distinguish an array of SARS-CoV-2 antibodies. For this, the research team will create a multiplexed vertical flow assay (xVFA) to simultaneously detect IgA, IgM, and IgG antibodies to the S protein (as well as variants of the S protein, such as delta, lambda, and other emerging variants), with separate immunoreaction sites dedicated to S-1, S-2, and the receptor-binding domain (RBD) of the S-protein in the SARS-CoV-2 virus and its most recent variants. Using existing and de-identified human serum samples, with the xVFA platform, the research team will screen COVID-19-positive samples, including those resulting from common variants (confirmed through reverse transcriptase-Polymerase Chain Reaction and sequencing) along with vaccinated samples and pre-pandemic un-vaccinated negative control samples. A neural network will then be trained using quantitative information from the multiplexed immunoreactions and the ground-truth clinical state over a set of remnant human serum samples. This training phase will (1) create a serodiagnostic algorithm to identify a positive immune response to SARS-CoV-2 infection (including common variants) or vaccination status using the multiplexed antibody measurements, and (2) identify the key subset of antibody-antigen interactions that most accurately represent and quantify an immune response to SARS-CoV-2 infection or protection via vaccination. A blinded testing phase will benchmark the performance enhancement of the multiplexed and data-driven approach to rigorously validate the trained inference network's generalization. By validating a new multiplexed vertical flow assay and serodiagnosis algorithm for COVID-19 immune protection, the research team aims to determine the significant improvements in sensitivity and specificity gained through the multiple measurements and computational analysis, which come with little added cost or operational steps, or required sample volume. This project will also establish a complementary educational outreach program that will involve (1) public interviews and popular science articles in news media and the internet; (2) undergraduate research opportunities involving underrepresented students; and (3) graduate student training through the organization of workshops, seminars and conferences.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.