基于PCA和神经网络的多形性胶质母细胞瘤驱动基因预测模型  

Identification of Cancer Driver Gene Model for Glioblastoma Multiforme Based on PCA and Neural Network Approaches

作　　者：周莉 常菁 郭珣 张艺 王志丹 

机构地区：[1]北京交通大学计算机与信息技术学院交通数据分析与挖掘北京市重点实验室,北京100044

出　　处：《现代生物医学进展》2017年第33期6553-6556,6592,共5页Progress in Modern Biomedicine

摘　　要：目的:通过对癌症基因表达数据的分析,预测多形性胶质母细胞瘤的驱动基因集。方法:基于主成分分析方法和神经网络,提出一种用于预测多形性胶质母细胞瘤驱动基因的系统生物学模型。首先对实验样本的原始表达谱数据进行预清洗,过滤掉无信息或表达不符合实验要求的表达数据,并对肿瘤表达谱数据进行标准化处理;然后对基因进行划分,相似突变率的基因将被划分到同一块中;最后通过学习神经网络,构建癌症相关基因的调控网络,得出驱动基因的预测集。结果:本研究应用上述模型,对多形性胶质母细胞瘤(glioblastoma multiforme,GBM)驱动基因进行预测。已发表的大量实验结果表明,我们预测出的大部分驱动基因在GBM中起重要作用。结论:我们提出一种对GBM表达谱数据分析的新方法,能够高精度地预测出该疾病的驱动基因,该模型同样能够较好地用于分析其它疾病的表达谱数据。Objective： Predicting a set of cancer driver genes by analyzing ganc expression data ofglioblastoma multiforme. Methods： We proposed a systematic approach to predicting driver genes for glioblastoma multiforme based on Principal Component Analysis and ~aining neural networks. First, the mwgene expression data were processed to filter out non-informative and low-expression data and then normalizing the tumor expression data. Second, we grouped the genes so that the ones with similar expression fold changes belong to the same group. Finally, in order to predict the driver gen~ set, we reconstruct cancer-related genes regulatory network through Neural network learning. Results： In our study, wc predicted a set of driver genes for glioblastoma multiforme, most of which played an impor- tant role in the carcinogenic process, as demonstrated by existing literatures. Conclusions： We proposed a new, general neural network model to analyze glioblastoma multiforme expression data, which is able to predict the set of drive genes for this disease with high accu- racy. The proposed approach could also be used to analyze the gane expression data of other important diseases.

关 键 词：驱动基因 基因表达谱 神经网络 系统生物学 主成分分析 

分 类 号：Q811.4[生物学—生物工程] R730.31[医药卫生—肿瘤]