基于隐马尔可夫模型的array-CGH数据贝叶斯分析

微阵列比较基因组杂化(comparative genomic hybridization,CGH)技术是用于发现DNA拷贝数变异的重要技术.本文根据DNA片段间的距离及测试样本与参考样本之间的荧光强度比,将微阵列实验的分辨率特征作为先验信息,建立用于分析微阵列CGH数据的贝叶斯隐马尔可夫模型.为解决基因数据变量多、抽样收敛速度慢的问题,本文将向前向后Gibbs算法应用到模型参数的马尔可夫链蒙特卡罗抽样估计中,以加快收敛.在对多形性胶质母细胞瘤基因数据分析中,本文方法能有效识别出DNA拷贝数异常的区域,所得结果与以往有关研究结论一致.模拟数据分析结果表明,在不同噪声下本文方法均能有效识别出异常区域,其误判率小于3%.

Bayesian analysis of micro array-CGH data based on hidden Markov model

The DNA copy number variant is related to cancers and genetic disease frequently. Micro array comparative genomic hybridization (CGH) is an important technique in finding this variant. Based on the distance between DNA fragments and the ratio of fluorescence intensities between the test and reference samples, we established a Bayesian hidden Markov model with array resolution prior to analyze micro array CGH data. Since the number of variables in a genic experiment is quite large, we use the forward-backward Gibbs simpler to accelerate the Markov chain Monte Carlo procedure. In applying our method to the gliobastona multiforme data, we detect the variants successfully and make inference matching up with the relevant tumor studies. In the simulated study, it is found that the method can detect all the variants under different noises. The error rates are all less than 3%.