Immunophenotyping by CyTOF and machine learning

我们的应用高维免疫表型使用细胞tof发现自然应用于临床设置。免疫学中测量多路复用的实验方法（例如SomaLogic的细胞因子阵列，单细胞RNAseq）已经出现爆炸式增长，我们关注的是CyTOF，因为它与免疫学问题中的生物学功能密切相关。我们的主要项目是开发新的机器学习衍生工具来自动分析CyTOF数据（与波士顿大学Pankaj Mehta的理论物理小组合作）。此次合作（部分由Moore基金会的种子基金资助）将Pankaj Mehta在机器学习方面的专业知识和Altan-Bonnet实验室在定量免疫学方面的专业知识协同起来，以优化CyTOF数据分析的新方法。我们强调免疫分类器的可解释性，以允许免疫学家利用我们的机器学习管道的结果进入可测试的假设和实验验证。我们正在与美国国立卫生研究院的临床实验室合作，应用我们的实验/理论管道来分析正在进行的临床试验的临床样本。我们与NIAID的Mike Lenardo小组合作，设计并验证了一种CyTOF面板，该面板为NIAID的临床基因组组提供外周血单个核细胞（PBMC）样本的一般免疫表型。这35个标签组涵盖了血液中所有主要的白细胞类型，并为针对疾病特异性表位的抗体的增强留下了开放的通道。我们正在分析Lenardo实验室研究的XMEN（“x连锁免疫缺陷伴镁缺陷、EBV感染和肿瘤”）和ALPS（“自身免疫性淋巴细胞增生性综合征”）患者的PBMC。初步测量表明，NIH临床基因组学分支内的大量可用样本如何与我们的高维表型相结合，以产生与疾病状态最佳相关的免疫分类器。我们将应用支持向量分类器来识别白细胞群，其频率变化和/或分化状态的变化与临床评分最相关。展望未来，我们将进一步合作，建立一个定制设计的实验和计算管道，以对多种原发性免疫缺陷患者进行稳健分类。因此，我们的目标是对我们的机器学习工具进行微调，以适应临床应用，同时通过获取原发性免疫缺陷患者的罕见样本，探索免疫稳态的全球破坏（如项目I所研究的）。类似的合作工作正在与NIAMS的Mariana Kaplan实验室进行。此次合作旨在加深我们对系统性红斑狼疮（SLE）患者中性粒细胞失调的理解。卡普兰实验室已经确定了一种新的低密度粒细胞群，这种粒细胞群会引发中性粒细胞诱捕陷阱（NETosis）的形成、IFN的分泌和血管损伤。我们的工作假设是，这种中性粒细胞功能障碍在系统性红斑狼疮中通过维持慢性炎症来改变体内平衡。在这里，我们用中性粒细胞特异性抗体扩展了我们的一般免疫表型细胞tof面板。中性粒细胞的分析需要处理新鲜的全血，卡普兰实验室每两周从国立卫生研究院医院的病人那里收到新鲜的样本。由于这个原因，我们基于细胞因子的免疫表型管道特别适合于识别SLE患者血液中免疫稳态的大规模破坏，因为我们已经验证了管道的稳健性，当新样本在很长一段时间（1个月）内积累、处理和分析时。因此，我们将收集具有不同临床表现的样本（n100），以便集中我们对SLE患者血液中动态平衡改变的理解。正如与Lenardo实验室的合作一样，我们正在利用与Kaplan实验室合作的机会，在临床环境中微调我们的CyTOF管道，并进一步探索如何在慢性炎症的背景下设置全局免疫破坏。

Our application of high-dimensional immune-phenotyping using CyTOF finds natural applications in clinical settings. There has been an explosion of experimental methodologies for measurement multiplexing in immunology (e.g. cytokine arrays by SomaLogic, single-cell RNAseq) and we are focusing on CyTOF for its close connection to biological function in immunological problem. Our main project has been to develop new machine-learning derived tools to automatically analyze CyTOF data (collaboration with Pankaj Mehta's theoretical physics group at Boston University). This collaboration (funded in part by a seed grant from the Moore foundation) synergizes Pankaj Mehta's expertise in machine learning and the Altan-Bonnet lab's expertise in quantitative immunology to optimize new methods in CyTOF data analysis. We emphasize interpretability of immunological classifier to allow immunologists to leverage the results of our machine learning pipeline into testable hypotheses and experimental validation. We are collaborating with clinical labs at the NIH to apply our experimental/theoretical pipeline to analyze clinical samples from ongoing clinical trials. Our collaboration with Mike Lenardo's group at NIAID led to the design and validation of a CyTOF panel that provides general immunophenotyping for peripheral blood mononuclear cell (PBMC) samples for the clinical genomics group at NIAID. This 35-label panel covers all the main leukocyte cell types of the blood and leaves open channels for augmentation with antibodies against disease-specific epitopes. We are profiling the PBMC of patients afflicted with XMEN ("X-linked immunodeficiency with magnesium defect, EBV infection, and neoplasia") and ALPS ("Autoimmune lymphoproliferative syndrome") under study within the Lenardo lab. Preliminary measurements demonstrate how the large number of available samples within the clinical genomics branch at NIH can be leveraged with our high-dimensional phenotyping to generate immunological classifier that best correlates with disease status. We will apply support-vector classifiers to identify the leukocyte populations whose variation in frequency and/or change in differentiation status best correlates with clinical scores. Moving forward, we will further this collaboration to build a custom-designed experimental and computational pipeline that robustly classifies patients from multiple primary immunological deficiencies. Hence, our goal is to fine-tune our machine learning tools to clinical applications, while probing the global disruption of immunological homeostasis (as studied in project I) with access to rare samples of patients with primary immunodeficiencies. Similar collaborative work is ongoing with Mariana Kaplan's lab within NIAMS. This collaboration aims at deepening our understanding of the dysregulation of neutrophils in Systemic Lupus Erythematosus (SLE) patients. The Kaplan lab has identified a new population of low-density granulocytes that trigger enhanced formation of neutrophil entrapment traps ("NETosis"), IFN secretion and vascular damage. Our working hypothesis for this collaboration is that such neutrophilic dysfunction in SLE globally displaces homeostasis by maintaining chronic inflammation. Here we expanded our general immune-phenotyping CyTOF panel with antibodies specific to neutrophils. Analysis of neutrophils requires the processing of fresh whole blood and the Kaplan lab receive fresh samples biweekly from patients at the NIH hospital. For this reason, our CyTOF-based immune-phenotyping pipeline is particularly well suited to identify large-scale disruption of immune homeostasis in the blood of SLE patients, as we have validated the robustness of the pipeline when fresh samples are being accrued, processed and analyzed over a large period of time (1 month). Hence, we will accrue samples (n100) with varied clinical presentations in order to focus our understanding of altered homeostasis in the blood of SLE patients. As in the collaboration with the Lenardo lab, we are taking the opportunity of collaborating with the Kaplan lab to fine-tune our CyTOF pipeline in clinical settings, as well as to further probe how global immunological disruption can be set in the context of chronic inflammation.