DMS/NIGMS 2: Collaborative Research: Developing Statistical Learning Methods for Revealing the Molecular Signatures of Microvascular Changes in Neural Injury

脊髓损伤(SCI)是一种创伤性和破坏性的疾病，可导致暂时或永久性瘫痪。脊髓损伤还会导致各种与瘫痪相关的疾病，这些疾病会使人虚弱，甚至危及生命。它还可以通过原发机械性损伤导致功能损害，继而在细胞水平上导致继发性损伤机制，包括细胞死亡和脊髓血管损伤。血脊髓屏障(BSCB)是调节血液和脊髓之间分子交换的结构，它的破坏是对功能恢复最不利的因素之一。BSCB至少由三种类型的单元间功能组成。了解这些细胞在损伤反应中的分子特征和功能是脊髓损伤研究的主要兴趣。PI将使用高通量单细胞RNA测序(scRNA-seq)，这是一种在单细胞分辨率下剖析基因表达的强大技术。他们还将开发新的现代统计学方法来阐明损伤脊髓中BSCB主要细胞类型的分子特征。这项研究将有助于我们了解脊髓损伤后血管反应的复杂性，为脊髓损伤的治疗产生新的治疗靶点，并为大数据分析创造新的统计机器学习工具。PIS计划将研究与教育和外展活动相结合，并向公众广泛传播结果、数据和软件。该项目旨在开发统计机器学习理论和方法，以回答关于识别在脊髓损伤中具有疾病相关功能的新的微血管亚群以及微血管与免疫细胞的串扰和通过渗透的免疫细胞进行调节的重要问题。其新颖之处在于将先进的scRNA-seq技术与强大的高维统计方法相结合，以实现三个生物学目标：a)定义微血管细胞的单细胞图谱；b)确定损伤脊髓中微血管细胞亚群变化的机制；c)确定微血管细胞和浸润性免疫细胞之间的串扰模式及其在神经炎症中的作用。这项研究产生的大量复杂数据将促使在开发可伸缩、稳健和可靠的统计工具以有效分析scRNA-seq大数据方面面临新的挑战。特别是，PIS计划开发(A)大规模细胞亚群学习工具，包括多尺度聚类和细胞标记搜索算法以及无监督的特征筛选和选择；(B)通过开发具有错误发现率控制的可扩展的Hodges-Lehmann方法，识别细胞标记和差异表达基因的高效和稳健的方法；(C)更有效的因素调节学习方法，通过识别重要的配体-受体对，利用基因的共同表达在亚群学习、细胞标记搜索、差异表达基因选择以及细胞与细胞之间的相互作用。这些新开发的方法将被应用于大规模的scRNA-seq数据，以回答SCI的生物学问题。该奖项反映了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。

Spinal cord injury (SCI) is a traumatic and detrimental condition that can result in temporary or permanent paralysis. SCI also causes various paralysis-related disorders that can become debilitating and often life-threatening. It can also lead to functional impairment via the primary mechanical injury followed by subsequent secondary injury mechanisms at cellular levels, including cell death and spinal blood vessel damage. Disruption of the blood-spinal cord barrier (BSCB), the structure regulating molecular exchange between blood and spinal cord, is one of the most detrimental factors to functional recovery. The BSCB is composed of at least three types of cell-to-cell functions. Understanding the molecular characteristics and functions of these cells in response to injury is a major interest of SCI research. The PIs will use high throughput single cell RNA sequencing (scRNA-seq), a powerful technique for the dissection of gene expression at single-cell resolution. They will also develop novel modern statistical approaches to elucidate the molecular characterizations of principal cell types of BSCB in the injured spinal cord. The study will help us understand the complexity of blood vessels in response to SCI, generate novel therapeutic targets for SCI treatment, and create new statistical machine learning tools for big data analysis. The PIs plan to integrate research with education and outreach activities and disseminate the results, data, and software broadly to the public. This project aims to develop statistical machine learning theory and methods to answer important questions regarding the identification of new subpopulations of microvessels that have disease-relevant functions in SCI, as well as microvessel crosstalk with and regulation by infiltrating immune cells. The novelty is to combine advanced scRNA-seq technologies with robust high-dimensional statistical methods for three biological aims: a) define single-cell profiling of microvascular cells; b) determine the mechanisms of the alteration of subpopulations of microvascular cells in the injured spinal cord; and c) identify the crosstalk patterns between microvascular cells and infiltrating immune cells and their roles in neuroinflammation. The large quantities of complex data generated from the study will prompt new challenges in developing scalable robust and reliable statistics tools for efficient analysis of big scRNA-seq data. In particular, PIs plan to develop (a) large-scale cell subpopulation learning tools including multi-scale clustering and cell marker hunting algorithms and unsupervised feature screening and selection; (b) efficient and robust methods for identifying cell markers and differently expressed genes by developing scalable Hodges-Lehmann's method with false discovery rate controls; (c) more efficient factor-adjusted learning methods that take advantages of co-expression of genes in subpopulation learning, cell marker hunting, differently expressed genes selection, as well as cell-cell interaction by identifying important ligand-receptor pairs. These newly developed methods will be applied to the large-scale scRNA-seq data to answer the biological questions for SCI.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.