FET: Small: Methods and Algorithms for microRNA Sensing: Interdependency Discovery and Inverse Problems

近年来的研究表明，一小群microRNA （miRNA）通常作为（疾病）生物标志物，其浓度从正常状态变化，可作为早期检测癌症、感染和心脏病等各种疾病的最有前途的方法之一。此外，已经做出了一些努力，将mirna作为应对医学治疗的预测工具，并将其本身作为治疗方法。此外，mirna驱动的信号级联在疾病的背景下起着至关重要的作用，其理解仍然是一个挑战。过去的一些研究表明，一组或集群中的mirna协同控制调控模式，特别是当它们共享特定的目标信使rna时。所有这些都强调了miRNA传感和发现miRNA-to-miRNA相互作用在大基因组范围内作为一个复杂的调控网络的重要性。然而，使用复杂的测序技术单独测量1000多个基因表达水平使得这种解决方案非常昂贵且难以解决。研究小组将开发一个由生物传感器和机器学习模型组成的集成框架，通过分析一小组低成本生物传感器的测量结果，恢复mirna浓度并发现它们在调节网络中的相互依赖结构。预计这项研究将在几个领域产生重大影响。1. 在技术和产品方面：了解miRNA-miRNA相互作用并能够监测miRNA表达水平可能有助于开发几种类型的癌症、心脏损伤、肌肉损伤和其他肌肉病变、糖尿病、肝损伤和许多感染性疾病的诊断工具。通过提供有效的miRNA传感和监测机制，该研究有可能降低医疗保健成本。2. 关于教育和学习：(i)培训研究生和本科生；（ii）扩大妇女和少数民族在这一领域的参与；（iii）传播研究成果；（iv）为k-12教师提供实习机会；(v)通过参加和组织多学科会议和讲习班加强对科学和技术的了解；（vi）与工业界和学术界建立合作努力；（vii）为miRNA传感而开发的解决方案可能的技术转让。提出的研究旨在建立一个由测量系统作为前端和机器学习算法作为后端组成的集成框架，以实现两个高层次的相互关联的目标：(i)开发机器学习解决方案的基础，该解决方案将分析来自一系列少量低成本生物传感器（其设计由所提议的机器学习框架指导）的测量结果，并发现大量mirna（例如，超过1000个mirna）中miRNA-to-miRNA相互依赖的结构；（ii）开发一个框架，通过利用miRNA-to-miRNA依赖结构，解决由所提出的传感器阵列从低维测量中恢复mirna分子浓度水平的逆问题。虽然研究的目标是生物学应用，但该研究将在几个方面推进理论和设计原则，并在许多其他应用中产生广泛的影响。具体而言，(1)本研究将首次研究和发展在间接低维观测下参数和非参数情景下概率图模型结构的学习理论。(2)本文还将引入一种基于图上密度演化的新范式，该范式可以利用高维信号的先验（依赖）结构来设计和优化用于高维信号恢复的压缩测量系统。(3)通过考虑高维信号中的某些结构（例如，由图形模型引起的条件独立性，稀疏性），所提出的工作将推进解决逆问题的理论和算法。(4)这项研究将首次导致廉价、模块化和快速作用的miRNA测量生物传感器阵列的发展，其设计原理与数据分析相结合并受其影响。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Recent studies have shown that a small group of microRNA (miRNA) often acts as (disease) biomarkers whose concentration changes from the normal state can be used as one of the most promising methods of detecting various diseases such as cancer, infection and heart diseases at early stages. Further, several efforts have been made to use miRNAs as a predictive tool in response to medical treatments, and as therapeutics themselves. Moreover, miRNA-driven signaling cascade plays a crucial role in the context of diseases, and its understanding remains a challenge. Several past works suggest that miRNAs in a group or cluster collaboratively control the regulatory patterns, especially when they share specific target messenger RNAs. All of these underline the importance of miRNA sensing and discovery of miRNA-to-miRNA interactions as a complex regulatory network in a large genome-wide scale. However, measuring 1000+ gene expression levels individually using sophisticated sequencing technologies renders such solutions very costly and intractable. The team of investigators will develop an integrated framework consisted of biosensors and machine-learning models for both the recovery of miRNAs concentrations and discovery of their interdependency structure in regulatory networks by analyzing measurements from a small group of low-cost biosensors. The research impacts are expected to be significant in several areas. 1. On technology and products: Understanding miRNA-miRNA interactions and being able to monitor miRNA expression levels is likely to contribute to the development of diagnostic tools for several types of cancer, cardiac damage, muscle damage and other muscle pathologies, diabetes, liver injury, and many infection diseases. By providing effective miRNA sensing and monitoring mechanisms, the research has potential to reduce the cost of health care. 2. On education and learning: (i) Training of graduate and undergraduate students, (ii) Broadening the participation of women and minorities in this field, (iii) Disseminating the research results, (iv) Providing internship opportunities for k-12 teachers, (v) Enhancing scientific and technological understanding by participation in and organizing multi-disciplinary conferences and workshops, (vi) Establishing collaborative efforts with both the industry and academia, and (vii) Possible technology transfer of the solutions developed for miRNA sensing. The proposed research aims at establishing an integrated framework consisted of a measurement system as a front end and machine-learning algorithms as a back end to achieve two high-level interrelated goals: (i) To develop the foundation for machine learning solutions that would analyze measurements from an array of small number of low-cost biosensors (whose design is guided by the proposed machine learning framework) and discover miRNA-to-miRNA interdependency structures in a large population of miRNAs (e.g., over 1000 miRNAs), (ii) To develop a framework for solving the inverse problem of recovering miRNAs' molecular concentration levels from a low dimensional measurement by the proposed sensor array, via leveraging miRNA-to-miRNA dependency structures. Although aimed at biology applications, the research will advance the theory and design principles in several fronts with a broad effect in many other applications. Specifically, (1) The research, for the first time, will investigate and develop the theory of learning the structure of probabilistic graphical models in both parametric and non-parametric scenarios under indirect low-dimensional observations. (2) It will also introduce a novel paradigm based on density evolution on graphs that can tap into the prior (dependency) structure of a high-dimensional signal to design and optimize a compressive measurement system for the high-dimensional signal recovery. (3) The proposed work will advance theory and algorithms for solving the inverse problem by taking into account certain structures in the high-dimensional signal (e.g., conditional independencies induced by graphical models, sparsity). (4) The research, for the first time, will lead to development of cheap, modular, and fast-acting array of biosensors for miRNA measurement whose design principle is integrated with and influenced by the data analytic counterpart.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.