融合功能元素网络的肿瘤异质性推断算法研究

A tumor is a heterogeneous population of cells containing a different complement of somatic mutations. Therefore, the effectiveness of therapeutic and prognosis are different between tumors. Even in the same tumor, their tumor cells are heterogeneous. Tumor heterogeneity is the main reason that causes tumor therapeutics failure. In this project, we aim to design new algorithms to infer subclones in tumor samples using prior biological knowledge so as to reveal the tumor heterogeneity. According to the evolutionary theory of tumor, a driver mutation is directly implicated as the cause of the tumor process—it confers a growth advantage to tumor cells and causes subclones occurrence. Compared with previous methods which use statistical sampling methods to fit the observed variant allele frequencies, our methods identify driver mutations and infer subclones by clustering the mutations having the similar mutation frequency to the driver mutations. We emphasize on designing quantitative index to measure the proportion of cells damaged by the mutations with respect to both various mutation types and characteristics of sequences. After that functional element networks will be constructed with prior knowledge. Then novel algorithms to identify driver mutation will be proposed by combining biological knowledge and functional element networks. Finally, novel clustering and heuristic methods will be introduced to infer subclone populations from tumor samples and to construct phylogenetic tree with the help of the identified driver mutations. Moreover a unified information access platform will be constructed on the base of biological data, research results and executable programs of this project. It will provide valuable information for biological and medical researchers. We will also employ our methods to analyze the heterogeneity of lung tumor in the area of Xuanwei and Fuyuan County, Yunnan province.

肿瘤异质性是目前肿瘤治疗方案失败的主要原因。肿瘤异质性是指同一肿瘤中存在有着不同基因型的细胞亚群。因此同一种肿瘤在不同的个体身上表现出不一样的治疗效果及预后，甚至同一个体身上的肿瘤细胞也存在差异。本项目将基于DNA序列数据，融合先验的生物知识，设计算法划分肿瘤样本中的细胞亚群来揭示肿瘤的异质性。我们的方法将跳出传统的通过统计采样来拟合数据的模式。依据肿瘤进化理论，从寻找导致肿瘤细胞亚群产生的驱动突变入手。重点研究通过综合考虑多种突变类型和序列特征，设计新的指标来衡量样本中发生突变的细胞含量。利用先验生物知识，构建功能元素网络。结合生物背景知识和功能元素网络，设计新的算法识别驱动突变。基于识别的驱动突变，设计聚类和启发式算法来推断肿瘤样本中的细胞亚群,构建细胞亚群之间的进化关系。最后搭建肿瘤异质性分析平台供生物医学研究人员使用。并且分析云南宣威、富源地区肺癌的异质性来指导该地区肺癌的防治。

肿瘤在不同的个体身上甚至同一个体身上都存在异质性，这也是目前肿瘤治疗方案失败的主要原因。驱动肿瘤发生发展的因素主要是由小部分的驱动突变或驱动基因引起的。因此本项目从识别癌症的驱动基因入手来研究癌症的异质性。考虑到细胞功能的系统性、复杂性我们综述了目前构建生物网络（如蛋白质网络）的计算网络，并对这些方法进行了可靠性评估。我们也提出了新的构建疾病相似性网络、动态蛋白质功能网络等多个生物网络的构造方法。基于构建的生物网络，融合与癌症相关的生物特性提出了一系列新方法来识别癌症驱动基因、关键基因以及非编码RNA。这些方法可以分为基于多网络扩散的方法、基于多网络融合的方法和基于网络加权的方法。我们还提出了机器学习的方法对跟癌症相关的基因特征进行提取和提出了新的集成学习方法通过集成多个弱分类器来提高癌症相关基因预测的准确性。本项目还研究了基于驱动基因的癌症亚型划分方法来研究肿瘤的异质性。最后研究中涉及的数据、研究成果以及算法实现都整合到一个统一的信息平台上来，方便相关人员使用。