基于高斯混合模型的肿瘤纯度估计

在癌症基因组学研究中，临床所得的肿瘤组织是由癌症和正常细胞组成的混合物，肿瘤不纯会对后续的数据分析产生严重影响。基于DNA甲基化的芯片数据，构造了一种简单的肿瘤纯度估计方法GmmPurify。首先借助公共正常样本，利用高斯混合模型定义了一个重要的统计量“信息贡献值”；然后筛选出具有高信息贡献值的DNA甲基化位点，构成差异甲基化位点集合；最后利用核密度方法估计肿瘤的纯度。将GmmPurify方法应用于9类肿瘤，得到的纯度估值与两类先进方法的结果高度一致。研究结果表明，在与肿瘤样本相匹配的正常样本缺失的情况下，借助公共正常样本，GmmPurify可以给出令人满意的肿瘤纯度估计。

Tumor purity estimation based on Gaussian mixture model

In cancer genomics or epigenomics research, tumor tissues obtained from clinic are mixtures of cancer and normal cells, and impure tumor may have a severe impact on subsequent data analyses. Based on DNA methylation microarray data, we propose a simple method, GmmPurify, for estimating tumor purity in this paper. First, we apply Gaussian mixture model on the common normal samples to derive an important statistics“information contribution value”. Then, we construct a set of differential methylation sites with high information contribution values and estimate their tumor purity by using kernel density method. To verify the performance of GmmPurify, we use it to compute the purities of nine types of tumors from The Cancer Genome Atlas (TCGA), and the obtained purity estimates are highly consistent with the results of two state of the art methods. The result shows that GmmPurify could provide a satisfactory tumor purity estimation in the absence of normal samples with match the current tumor samples.