一个基于知识驱动策略识别肿瘤非编码驱动突变的新方法

Driver mutations in cancer genome is the primary driving force for tumor growth and development. Coding driver mutations may affect the biochemical functions encoded by a cancer gene, while non-coding mutations may interfere with the expression pattern of the cancer genes. Currently, the number of non-coding driver mutations discovered is far less than that of coding driver mutations. A possible reason is because of the approach used for detecting non-coding driver mutations, which focuses on evaluating the mutation frequency of a specific regulatory region. However, a gene may be regulated by many regulatory elements, each of which may be mutated and confers the same functional effect, consequently lowering the possibility of discovering one regulatory region with significant frequent mutations. In this proposal, we propose a novel knowledge-driven approach. It starts from predicting epi-driver genes that are abnormally expressed and confer growth advantage to tumor cell by using a combined computational tool. Once epi-driver genes are predicted, further prediction is made to predict regulatory elements including those located distal to the gene, and a group of regulatory elements is formed. Then, mutation frequency is computed and evaluated based on this group of regulatory elements, rather than on specific ones, as done before. Groups of regulatory elements with potential growth advantage are predicted based on their integrated mutation frequency relative to carefully selected backgrounds. Finally, annotations such as transcription factor binding sites will be used to filter out likely non-coding driver mutations. Compared to current methods that focus on one regulatory region, our proposed method is of great advantage for finding more non-coding driver mutations.

肿瘤基因组的驱动突变是肿瘤细胞演化的主要驱动力。编码区域的驱动突变可影响关键基因编码的蛋白的功能，而非编码区域的驱动突变则可影响关键基因的表达模式。目前发现的非编码驱动突变的数量要远远小于编码驱动突变。一个可能的原因是由于现有方法大多都基于对固定调控区域上的突变频率的评估。而一个基因可能受包括多个元件的调控，不同元件上的突变可能会达到影响关键基因表达的同样效果，从而降低了突变发生在特定区域上的概率。为此，本项目计划采用一个创新的知识驱动的策略，首先预测在肿瘤细胞中基因表达异常且可能对肿瘤生长有促进作用的驱动基因，然后通过计算预测等手段构建该基因的调控元件集合，并以元件集合为对象，来检测其整体突变频率并与背景频率比较，来预测有潜在驱动性的调控元件集合。最后，再结合转录因子结合位点等注释信息来推测非编码驱动突变。与基于固定区域的评估方法相比，该方法可显著提高识别非编码驱动突变的效率。

本项目从预测肿瘤细胞中表达异常的驱动性基因—表观驱动基因出发，结合DNA甲基化修饰组学数据，在表观驱动基因的启动子区域，筛选获得了与其所在基因表达模式又协同性修饰变化的甲基化位点，并推测这些非编码位点的甲基化修饰可能是驱动肿瘤生长的驱动性事件。在此基础上，通过对这些甲基化位点进行进一步筛选，我们在多个肿瘤中建立了基于所筛选的甲基化位点的肿瘤诊断和预后的模型，探索和验证了基于知识驱动的模式来寻找肿瘤标志物的可行性。此外，本项目也探索了单细胞数据的生物信息学分析等研究方向。在本项目的支持下，我们在Nucleic Acids Research, Cell Reports, Cell Discovery, Bioinformatics等国际顶尖专业期刊上发表了7篇有本项目标注的SCI论文，给后续的临床应用提供了新的研究方向。