基于高维短序列生物数据的系统重构研究

One of systems biology's main purposes is to understand the mechanism of various biological phenomenon from a systematic view. Thus how to use the large amount of experimental data especially the time evolution data to reconstruct the biological system is an important method to understand biological systems. Though various kinds of time series analysis methods have been fruitfully proposed, the traditional methods mainly focus on the analysis of low dimensional long-term data, and several algorithms such as prediction, causality detection and system reconstruction has successfully proposed. However, there are still few effective methods to handle high dimensional short-term data such as high throughput data developed these years. Therefore, this proposal will aim to develop algorithms particularly effective for high dimensional short-term time series data. We will include theoretical analysis, algorithm design and validation by biological experiments. It is expected to investigate how to transfer the information embedded in high dimensional data into time domain such that we can reconstruct the dynamics of the system. Also it is expected to investigate the causality theory based on the map between reconstructed attractors so that we can reconstruct the regulation network. And it is expected to investigate the identification for large amount of system parameters so that it is possible to make system reconstruction based on short measurements. Finally, we will apply the developed technique to investigate the high throughput data for diabetes, in order to validate and make further exploration on the mechanism of the disease.

系统生物学的一个重要目标是从系统的角度来认识各个层次的生物学现象，而探索和理解生物系统的一个重要手段是通过分析海量的各层次多种类的实验数据特别是生物系统的时间序列数据来重构系统的各种动力学行为。目前已有的数理方法主要集中于对低维长时间序列数据的动态行为研究，而对于以基因芯片等高通量生物数据为代表的高维短时间序列数据还缺乏相应的分析和重构手段。本项目通过理论研究、算法设计及结合实验结果，旨在获得针对高维短时间序列数据的系统重构方法，主要包括研究高维短时间序列数据的可靠预测和历史追溯以实现数据的动力学重构；研究短时间序列变量间的因果关系检测以实现系统调控网络的重构；研究基于短测量数据的大规模参数识别算法以实现系统模型的重构；并应用以上理论算法来研究II型糖尿病高通量数据的信息挖掘以验证和探索疾病的机制和治疗方法。

如何从系统生物学的大量数据中挖掘出系统有用信息，如重构出系统的动力学模型，推断出因果关系并和重构出调控网络，辨识各层面各类别的参数，是探索和理解生物系统的一个重要手段和有意义的数理问题。本项目主要针对高维短时间序列数据，研究生物系统的重构方法，旨在开发出一套基于非线性科学的通用方法，并用于解决实际问题。.. 在本项目执行期间，针对项目计划中计划的三个主要问题，即系统预测，因果推断及网络重构，参数辨识及模型重构，分别展开了相应的研究并获得了预期的成果。对于系统预测，我们开发一套基于嵌入定理的逆向嵌入方法，特别针对高维短数据提供了预测的可能；对于因果推断，我们发展了近期国际上的研究热点，即基于交叉映射的因果推断方法，开发了光滑交叉映射检测方法，将检测所需要的时序列最短长度从至少O(10^3)大幅降低到O(10)，成果获得了很多国际同行的关注；对于参数辨识与模型重构，我们提出了未知不稳定点的随机检测方法，克服了传统方法需要预先知道不动点种类的检测条件，可以通过一套统一的方法将所有种类的不动点同时稳定化并检测出来，并刻划出稳定化后的不同不动点各自的吸引域。上述三个主要问题的研究成果分别发表在三个SCI杂志上，目前还有两篇论文正在投稿评审中。