面向进化基因组学的高通量测序数据流形建模

Evolutionary genomics is a hot topic in computational biology, which focuses on discovering the adaptation mechanism of life organisms to environmental changes at molecular biology level. In this project we propose a novel idea of adopting manifold modeling to precisely exhibit the details in gene evolution, and correctly explore the topology structures of gene sequencing data in the sequence space as well as their clustering formations, and accurately recover the gene evolutionary path concealed in next-generation sequencing data. Based on our previous works, we propose an efficient method of constructing the k-nearest neighbor list for data in non-Euclidean space rapidly with the aid of pseudo-metric space partitioning tree, and a detail-preserving sampling method based on topological similarity, which circumvents the computational difficulties of applying manifold learning to large-scale data. Moreover, we introduce the concept of constrained clustering and sub-space clustering into manifold modeling, which makes use of a small number of labelled data to determine a manifold subspace that reflects specially-defined classification significance. In this way the dimensions in the manifold space introduced by meanful variations are accurately discriminated with those introduced by random noises. Moreover, by constrained clustering the the natural borders of the data set in the manifold subspace are explored, as well as the structure details of gene evolutionary path. The implementation of the project will lead to a powerful tool for genomics analysis and provide insights to development of new methods for manifold modeling.

进化基因组学分析是计算生物学的热点领域之一，是在分子生物学层面揭示生物对环境适应机制的主要方法。本项目拟提出将流形建模引入进化基因组学分析以准确描绘生物基因进化过程细节的新思路，正确提取并展示基因序列数据在空间的拓扑结构和群聚关系，准确复原高通量测序数据所反映的基因进化路径。本项目在申请人前期工作的基础上，拟提出运用伪度量空间分割树的非欧氏空间快速k近邻列表构造方法，和基于拓扑相似性的细节保真抽样方法，解决流形建模技术应用到大规模数据分析的计算负荷问题。更进一步，将约束聚类和子空间聚类的思想引入流形建模，利用少量标定样本确定具有特定分类意义的流形子空间，准确区分流形空间的有效维度和干扰维度，寻找数据集在流形空间上的自然分类边界，准确复原基因进化路径的结构细节，为基因组学研究提供有力的分析工具，同时为发展流形建模的新方法提供思路。

进化基因组学分析是计算生物学的热点领域之一，是在分子生物学层面揭示生物对环境适应机制的主要方法。本项目提出将流形学习和深度神经网络引入进化基因组学分析以准确描绘生物基因进化过程细节的新思路，正确提取并展示基因序列数据在空间的拓扑结构和群聚关系，准确复原高通量测序数据所反映的基因进化路径，实现准确的物种序列分类。本项目在申请人前期工作的基础上，提出了运用伪度量空间分割树并行快速级联聚类方法，和基于动态地标选择技术的分治式并行级联聚类方法，解决流形建模技术应用到大规模数据分析的计算负荷问题。更进一步，引入深度神经网络和流形聚类的思想，利用少量标定样本确定具有特定分类意义的流形特征子空间，寻找数据集在流形空间上的自然分类边界，准确复原基因进化路径的结构细节，提升了物种基因序列分类的准确度，为基因组学研究提供了有力的分析工具。相关成果已发表研究论文7篇，其中SCI论文5篇，EI论文2篇；其中JCR一区的SCI国际著名期刊4篇（含中科院分区一区的PLOS Computational Biology, Bioinformatics顶级期刊论文各1篇）；申请发明专利5项，授权发明专利2项。超额完成项目目标。