大规模高分辨质谱数据挖掘新方法研究

High-resolution mass spectra(HRMS) plays an important role in structure elucidation. However, the mining of discriminant markers from large-scale GC-MS or LC-MS dataset and the identification of them via HRMS are still difficult for most researchers. Presently the markers are often discovered by manual preprocessing and pattern recognition, then identified by searching MS libraries. This procedure is time-consuming and subjective, and the spectra in MS libraries are limited. So some novel methods for preprocessing, pattern recognition and identification are needed urgently. In this project, we will implement baseline-fittig, peak detection, automatic deconvolution and alignment methods to construct 2D matrix for pattern recognition and corresponding HRMS for structure elucidation based on high performance computing techniques. Then random forests or sparse linear discriminant analysis will be employed to discover the influential markers effectively. For the markers not including in the MS libraries, accurate m/z values, mass spectra calibration, isotopic abundance, PubChem database, retention index and in silico fragmentation will be adopted for molecular formula and structure identification. This study can provide a novel and systemic platform for analyzing and mining HRMS dataset of complex system, which is meaningful to several research areas such as metabolomics, food safety, active compounds of herbal medicine and etc.

高分辨质谱在结构鉴定中起着非常重要的作用，但是从大规模的气质或液质数据中通过预处理和模式识别挖掘出有判别能力的标记物，然后利用高分辨质谱对其鉴定仍是复杂体系分析的关键和难点之一。目前主要采用预处理方法与模式识别，鉴定则依赖于库检索。目前预处理方法耗时且主观性大以及谱库覆盖范围有限，因此需要新的预处理、模式识别与鉴定等方法。本项目在高性能计算平台支撑下，实现高分辨质谱自动基线校正、峰检测、多元分辨和校准等方法，可快速从联用数据中挖掘用于模式识别的二维矩阵；采用随机森林和稀疏线性判别分析等方法识别出标记物；对无法用质谱库进行鉴定的标记物，通过高分辨质谱精确质量、质谱校准、同位素丰度、PubChem数据库、保留指数以及理论裂解规律等方法进行定性分析。项目成功实施将为复杂体系高分辩质谱数据提供更好分析与挖掘方法，对目前几个研究热点，如代谢组学、食品安全、天然药物活性成分等领域有很强的现实意义。

高分辨质谱在复杂体系分析中起着非常重要的作用，但是从大规模的高分辨质谱数据中通过预处理和模式识别挖掘出有判别能力的标记物，然后利用高分辨质谱对其鉴定仍是复杂体系分析的关键和难点之一，大规模高分辨质谱数据挖掘仍非常具有挑战性。因此需要开发新型化学计量学算法准确高效地从大规模原始联用质谱数据中挖掘出有意义的信息。本项目首先搭建可存储及分析太字节(Terabyte)数据的硬件平台，包括30TB存储空间的网络存储服务器及12TFLOPS单精度浮点计算能力的多路GPU计算节点，并针对联用质谱数据中的纯离子色谱峰提取、基线校正、峰检测、峰校准以及高分辨质谱定性等问题进行了系统性的研究或探讨，提出了K-均值纯离子色谱提取方法（K-means clustering Pure Ion Chromatogram, KPIC）、自动双边指数基线校正算法（Automatic Two-side Exponential Baseline correction algorithm，ATEB）、小波空间多尺度峰值检测（Multiscale Peak Detection, MSPD）、递归小波空间峰检测算法（Recursive Wavelet Peak Detection，RWPD）、移动窗口快速傅立叶交叉相关（Moving Window Fast Fourier Transfrom cross-correlation, MWFFT）、提出质谱辅助信号分辨技术（MS-Assisted Resolution of Signals，MARS）以及新型并行分子式生成器(Parallel Formula Generator, PFG)。所提出的化学计量学方法成功用于化工园区的挥发性有机化合物（Volatile Organic Compounds, VOCs）在线源解析以及男性不育代谢组学诊断研究。