差异甲基化分析和抗癌药物敏感性预测中的计算模型

发布时间：2019-05-16 15:58

【摘要】：肿瘤细胞与正常细胞的差异甲基化分析是癌症表观遗传学研究的重要内容之一。肿瘤组织往往由肿瘤细胞、正常细胞、侵入免疫细胞等多组织混合而成,其中正常细胞的混入对随后的差异甲基化分析有较大影响,但目前尚未有差异甲基化分析的工作考虑到肿瘤细胞纯度的影响。基于多种基因组学数据预测抗癌药物的敏感性是实现肿瘤个体化医疗的核心步骤。目前,针对药物敏感性预测问题的研究大多利用各种基因组学数据对药物敏感值进行稀疏回归分析,对于单个药物逐一建模。此类方法忽略了药物自身分子特征对药物敏感性的影响,对大部分药物的预测准确度难以达到临床应用的要求。本论文对上述两个计算生物学问题进行了系统的研究。对于肿瘤细胞纯度估计问题,我们首先利用了肿瘤与正常组织中差异最显著的若干位点进行了密度估计,建立了基于这些位点的纯度估计模型。预测的结果与基于拷贝数、基因表达和二代测序得到的结果高度一致。基于上述纯度估计的结果,我们利用广义线性模型建立了考虑到肿瘤纯度的肿瘤-正常差异甲基化位点估计模型,针对肺腺癌、结直肠癌的差异甲基化分析表明,考虑到肿瘤纯度的方法比原有的秩和检验在差异位点个数、肿瘤间统计量一致性等指标上表现更优。对于抗癌药物敏感性预测问题,我们首先研究了基因组特征相似的细胞系对于同一个药物,以及化学相似的药物对于同一个细胞系之间药物敏感性的相似性。随后构建了细胞系-药物双层网络模型,并提出局部线性方法预测抗癌药物对于细胞系的敏感性。预测结果显示,相比现有的“弹性网络回归”等稀疏回归方法,双层网络方法利用更少的模型参数和特征信息,却具有更高的预测准确度。我们也对”癌症基因组计划”(CGP)中缺失的敏感性数据进行了补充,结果表明BRAF突变的细胞系对三个MEK抑制剂敏感性更高,这与实验得到的结果一致。
[Abstract]:Differential methylation analysis between tumor cells and normal cells is one of the important contents in the study of cancer epigenetics. Tumor tissue is often composed of tumor cells, normal cells, invading immune cells and other tissues, in which the mixing of normal cells has a great influence on the subsequent differential methylation analysis. However, no differential methylation analysis has taken into account the effect of tumor cell purity. Predicting the sensitivity of anticancer drugs based on a variety of genomics data is the core step to realize individualized medical treatment of tumors. At present, most of the studies on the prediction of drug sensitivity use a variety of genomic data to carry out sparse regression analysis of drug sensitivity, and model a single drug one by one. This method ignores the influence of drug molecular characteristics on drug sensitivity, and the prediction accuracy of most drugs is difficult to meet the requirements of clinical application. In this paper, the above two computational biology problems are studied systematically. For the problem of tumor cell purity estimation, we first use some of the most significant differences between tumor and normal tissues to estimate the density of tumor cells, and establish a purity estimation model based on these sites. The predicted results are highly consistent with those based on copy number, gene expression and second generation sequencing. Based on the results of the above purity estimation, we established a tumor-normal differential methylated site estimation model considering tumor purity by using the generalized linear model. The differential methylation analysis for lung cancer and colorectal cancer showed that, The method considering tumor purity is better than the original rank sum test in the number of differentially located points and the consistency of statistics between tumors. For the prediction of anticancer drug sensitivity, we first studied the similarity of drug sensitivity of cell lines with similar genomic characteristics to the same drug and chemical similar drugs to the same cell line. Then the cell line-drug double-layer network model was constructed, and a local linear method was proposed to predict the sensitivity of anticancer drugs to cell lines. The prediction results show that compared with the existing sparse regression methods such as "elastic network regression", the double-layer network method makes use of less model parameters and characteristic information, but has higher prediction accuracy. We also supplemented the missing sensitivity data in Cancer Genome Project (CGP). The results showed that BRAF mutant cell lines were more sensitive to three MEK inhibitors, which was consistent with the experimental results.
【学位授予单位】：上海师范大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：R96

【相似文献】