CAREER: Machine learning and signal processing methods for analyzing single-cell sequencing data

高通量基因组测序技术，也被称为下一代测序技术，已经给基因组学和医学带来了革命性的变化。这一成就主要归功于使用计算方法从新的测序技术产生的大量基因组小片段中提取信息。最近，对单个细胞进行测序，即所谓的单细胞测序，显示出比传统的大量细胞测序，即称为整体细胞测序，有了显著的进步。这是因为单细胞测序能够在细胞水平上发现新的生物学知识，并更好地了解单个细胞的功能，而这是通过整体测序无法获得的。近年来发展迅速的单细胞测序技术在微生物学、神经生物学、免疫学、发育生物学等领域产生了重大影响，引起了人们的极大兴趣。随着单细胞技术的快速发展，单细胞测序数据及其应用不断增长，而用于分析这些数据的计算方法却落后了。与批量测序相比，单细胞测序在数据分析方面带来了新的挑战，因为单个细胞的DNA和RNA数量很少，而且单细胞测序过程中需要额外的步骤。此外，在任何给定的实验中，数千或数百万个细胞被并行测序，导致需要分析的海量数据集。该项目将为分析单细胞测序数据奠定坚实的计算基础。该项目的成果将有助于推进生物科学和改善人类健康，通过洞察需要单细胞解析的关键生物未知因素，如癌细胞的进化和干细胞的发展。该项目还将有助于高需求和多学科领域的生物信息学和计算基因组学的教育和培训。使用先进的计算方法进行数据分析在从单细胞测序数据中提取准确和有意义的信息方面发挥着至关重要的作用。目前的单细胞测序数据分析方法是从批量测序技术改编而来的。然而，目前的方法并不是为了应对单细胞测序数据分析中的新挑战而设计的，例如广泛的噪声、零膨胀和丢失数据、不均匀的基因组覆盖、数据多模式和大量数据。该项目旨在通过开发基于信号处理和机器学习技术的新计算方法和算法来解决单细胞测序数据分析方面的新挑战。这项研究的重点是利用DNA单细胞测序数据识别拷贝数变异形式的基因组变异，并使用RNA单细胞测序数据对细胞进行聚类。开发的方法和算法将通过(I)通过应用稀疏优化提供基因组覆盖数据的最佳表示，(Ii)通过在信号处理中使用去噪方法来建模和降低噪声，(Iii)通过应用数据驱动的学习模型来探索跨细胞的信息，以及(Iv)通过采用网络和单词嵌入模型来整合先验知识，从而显著提高从复杂和海量单细胞测序数据中提取准确信息的知识。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。

High throughput genome sequencing technologies, also known as next generation sequencing technologies, have revolutionized genomics and medicine. This achievement is mostly owed to employing computational methods to extract information from massive numbers of small fragments of genomes resulting from the new sequencing technologies. More recently, sequencing individual cells, known as single-cell sequencing, shows significant advancements over conventional sequencing of a large population of cells, called bulk cell sequencing. This is because single-cell sequencing enables discovery of new biological knowledge at the cellular level and a better understanding of the function of an individual cell, which cannot be obtained via bulk sequencing. The emerging and fast-growing single-cell sequencing technology has attracted much interest and has had major impacts on several fields, such as microbiology, neurobiology, immunology, and developmental biology. With rapid advances in single-cell technologies, single-cell sequencing data and their applications continue to grow, while computational methods for analyzing these data are lagging behind. Compared to bulk sequencing, single-cell sequencing introduces new challenges in data analysis due to the low amount of DNA and RNA from a single cell and the extra steps in the sequencing process for single cells. Moreover, thousands or millions of cells are sequenced in parallel in any given experiment, leading to massive data sets to analyze. This project will build a strong computational foundation for analyzing single-cell sequencing data. The project outcome will contribute to advancing biological sciences and improving human health by providing insight into critical biological unknowns requiring single-cell resolutions, such as evolution of cancer cells and development of stem cells. The project will also contribute to education and training in the high in-demand and multidisciplinary field of bioinformatics and computational genomics.Data analysis using advanced computational methods plays an essential role in extracting accurate and meaningful information from single-cell sequencing data. Current single-cell sequencing data analysis methods have been adapted from bulk sequencing technologies. The current methods, however, are not designed to cope with the new challenges in single-cell sequencing data analysis, such as extensive noise, zero inflation and missing data, non-uniform genome coverage, data multimodality, and large amount of data. This project aims to address the new challenges in analyzing single-cell sequencing data by developing novel computational methods and algorithms based on signal processing and machine learning techniques. The focus of this research is on identifying genomic variations in the form of copy number variations using DNA single-cell sequencing data, and clustering cells using RNA single-cell sequencing data. The developed methods and algorithms will significantly advance knowledge in extracting accurate information from complex and massive single-cell sequencing data by (i) providing optimal representation of genome coverage data by applying sparse optimization, (ii) modeling and reducing noise by employing denoising methods in signal processing, (iii) exploring information across cells by applying data-driven learning models and (iv) incorporating prior knowledge by adapting network and word embedding models.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.