中文摘要：

目的探讨网络解卷积算法对网络结构的优化效果。方法模拟研究采用四种网络算法对具有金标准的DREAM5平台数据进行网络构建，并评价解卷积优化前后的网络准确性。实例研究使用RF回归对卵巢癌晚期化疗敏感性患者的基因表达数据构建网络，再通过网络解卷积算法优化。结果模拟研究结果表明，四种网络构建方法推断出来的网络结构在解卷积算法优化后，其准确性均有不同程度的提高，其中基于线性相关的网络构建方法提高幅度明显大于CLR和剧-算法；实例分析结果表明，采用RF-ND方法构建的网络移除了部分间接边，其优化后能得到与现有数据库较为一致的网络结构。结论应用解卷积算法能够优化不同网络构建方法得到的网络，实际中能得到准确度较高的网络结构。

英文摘要：

Objective To investigate the performance of the network optimization based on network deconvolution. Methods In simulation studies, we performed four network reconstruction methods to construct the gene regulatory network on the data from DREAM5 platform which have contained the gold standard. Then we compared the accuracy of before and after optimization based on network deconvolution algorithm. In pritical studies, we applied random forest regression to construct an original network on gene expression data which comes from the advanced ovarian cancer patients that was susceptible to chemical therapy. Finally, we performed the network deconvolution method to optimize the structure of it. Results Simulation studies demonstrated that the accuracy of networks that reconstructed by four methods was increased to some degree. For the range of improvement, method that based on linear correlation was greater than CLR and RF. In practice, the method based on RF-ND removes some indirected edges and achieves satisfactory network structure that consistent to the existing database. Conclusion The algorithm of network deconvolution could optimize the structure of network constructed by the different methods and obtain the network with higher accuracy.