基于动态Bayesian网络的基因调控网络建模

Dynamic Bayesian network approach for modeling gene regulatory networks

为了精确建模与推断基因调控网络,提出一种基于动态Bayesian网络的多数据融合方法(SP-DBN)。该方法利用结构期望最大算法进行未知结构学习,基于粒子滤波方法完成参数学习,可有效处理数据缺失与噪声问题,更好地捕捉数据中固有的动态特性,并通过其先验结构,在基因表达数据的基础上,自然地融合转录因子绑定位点等多数据源信息。基于酿酒酵母的真实数据,实验结果表明:对于仅采用基因表达数据的情况,SP-DBN的敏感度与特异度分别提高到19%和95%;融入绑定位点数据后,SP-DBN的敏感度可从19%进一步提升至20%,而特异度则仍保持在95%的水平。

A dynamic Bayesian network-based multiple data fusion method was used to improve the modelling accuracy and the inferred gene regulatory networks. Structural expectation maximization and particle filtering are used to learn the unknown network structure and the parameters in a method that can effectively handle missing and noisy data. The method captures the dynamic nature of the biological system and naturally incorporates other data from transcription factor binding location data into the original gene expression data. The effectiveness of the method is shown by tests on Saccharomyces Cerevisiae cell cycle data. The results show that the sensitivity and specificity of the method are increased by 19% and 95% for the gene expression data itself and the prediction accuracy is raised further with multiple data sources.