基于熵值的冠心病基因网络模块划分方法评价与模块功能相似度分析

Comparison of different methods of module division by entropy and functional similarity of gene network and its modules for coronary heart disease

目的解决网络模块划分方法优选问题,并对划分结果进行功能解析和评价。方法利用多种常用模块划分方法,包括MCLi QUE,Cluster one,NCMine,PEWCC,Fuzzifier cluster,Community cluster,Connected component cluster,MCL cluster,MCODE cluster,Spectral clusters of protein sequences和AP等,对冠心病基因网络进行模块划分,通过网络结构熵值计算进行评价,优选出一种方法进行划分模块。基于基因功能富集分析对模块的通路功能进行富集,并利用杰卡德相似指数分析方法,对模块与总网络的功能进行比较和分析。结果用MCODE cluster方法划分出11个模块,网络熵值为4.33637,在几种方法中熵值最小,最适于冠心病基因网络模块的识别。基于基因功能富集分析和杰卡德相似指数分析方法,发现11个模块能涵盖冠心病相关通路中的38条,覆盖率达到73.1%。其中,模块3,4和7能用较少的基因富集到较多的功能。结论网络信息熵可以为模块划分方法的评价提供一种可行方案。划分后的模块可以代表原疾病网络的大部分功能,并在功能富集上有一定优势,对进一步简化和理解疾病网络具有实际意义。

OBJECTIVE To solve the problem with optimization of the network module division method and to analyze and evaluate the results of the division. METHODS A variety of commonly used module division methods were used to segment the gene network of coronary heart disease（CHD）,including MCLi QUE, Cluster one, NCMine, PEWCC, Fuzzifier cluster, Community cluster, Connected component cluster, MCL cluster, MCODE cluster, Spectral clusters of protein sequences and AP. The effect of division was evaluated by the network structure entropy. Then, based on the gene function enrichment analysis, the functional pathways of CHD gene network and its modules were enriched and compared by Jekard′s similarity index. RESULTS It was found that the MCODE cluster method divided11 modules with the lowest network entropy of 4.33637, suggesting that MCODE cluster method was the most suitable for the network module division of CHD. Based on gene functional enrichment analysis and Jekard′s similarity index analysis, it was found that the 11 modules could cover 38 cases of CHDrelated pathways with coverage of 73.1%. Among them, the modules 3, 4 and 7 could enrich more functional pathways with fewer genes. CONCLUSION The network information entropy can provide a feasible scheme for the evaluation of the module division method. The partitioned modules can represent most of the functions of the original disease network and have some advantages in functional enrichment,which can help further simplify and understand the disease network.