基于GPU-CPU混合并行计算的微生物组大数据索引与搜索方法学研究

Metagenomics is one of the most important and widely used approaches to study microbiome, which has enormous potential in solving crucial issues such as human health, environment protection and bio-energy. Although the volume of metagenome datasets has been increasing exponentially, the paucity of efficient methods for the “microbiome to microbiomes” type of search has greatly hindered the in-depth annotation of an unknown microbiota based on the vast amount of existing metagenomic data. The key technological bottlenecks here include the difficulty of reducing searching complexity among massive number of samples and the inability to comprehensively, rapidly yet precisely measure the structural or functional ‘matches’ between communities. To tackle these challenges, in this work we propose to establish a normalized indexing strategy based on microbial community features to improve the speed of search response, and then develop a multi-angle metagenome-to-metagenome alignment algorithm using the GPU-CPU hybrid parallel computing to enable the whole-microbiome oriented rapid search. Such a strategy should enable rapid and accurate identification of the microbiome in the database that is of the highest similarity in either community structure or function to the query microbiome, so that the associated metadata of the matching microbiome(s) can be used to annotate the query microbiome or to predict the environmental/host properties associated with the query microbiome. Based on this novel methodology, the first large-scale microbiome search engine that allows both ‘structure-based’ and ‘function-based’ search of metagenomes will be demonstrated using oral microbiome as a model, which will serve the research by enabling “gaining new knowledge by reviewing old” in the emerging area of Microbiome Data Science.

元基因组学是研究微生物组最重要和最普及的手段之一，在解决健康、环保、生物能源等重大问题上有巨大潜力。目前，元基因组学数据正在呈指数级增长，然而“群落对群落”式的微生物组搜索尚无高效解决方案，严重阻碍了基于海量已知数据对新样本的深入解读。此中的技术难点包括如何降低海量样本中的搜索复杂度以及如何全面、迅速、准确地衡量群落间的匹配性等。针对以上瓶颈，申请人提出建立微生物群落特征的归一化索引，以提高搜索响应，并开发基于GPU-CPU混合并行计算的多角度群落对比算法，能够在急速增长的已知数据中，迅速准确地找到与新样本在群落结构或功能上高度相似的匹配，从而根据后者的相关信息来注释新样本的特性并预测其所代表的环境特征。在此基础上，将以口腔微生物组为模式，示范首个“结构”和“功能”搜索兼备的大规模微生物组数据引擎与共性计算分析工具，服务于微生物组的“温故而知新”。

元基因组学是研究微生物组最重要和最普及的手段之一，在解决健康、环保、生物能源等重大问题上有巨大潜力。目前，元基因组学数据正在呈指数级增长，然而“群落对群落”式的微生物组搜索尚无高效解决方案，严重阻碍了基于海量已知数据对新样本的深入解读。此中的技术难点包括如何降低海量样本中的搜索复杂度以及如何全面、迅速、准确地衡量群落间的匹配性等。针对以上瓶颈，本研究开发了微生物组搜索引擎，通过建立微生物群落特征的归一化索引，以提高搜索响应，并开发基于并行计算的多角度群落对比算法，在急速增长的已知数据中，迅速准确地找到与新样本在群落结构或功能上高度相似的匹配，从而根据后者的相关信息来注释新样本的特性并预测其所代表的环境特征。在此基础上，将以人体微生物组为模式，示范首个“结构”和“功能”搜索兼备的大规模微生物组数据引擎与共性计算分析工具，服务于微生物组的“温故而知新”。