基于单样本边扰动网络的癌症个体化诊疗研究

Precision medicine is emerging as a key paradigm for cancer treatment, which aims to treat cancers in accordance with their distinct characteristics. Compared to the traditional ways of classifying cancers based on their mutation patterns, gene network-based characterization of a cancer can capture more accurately the true essence of each disease. In this project, we propose to use differential co-expression information of expressed genes in a cancer to represent the disease. We will first build a co-expression network for samples at the same TNM stage for the target cancer type and provided transcriptomic data, and then stage each new sample based on the similarity between the co-expression patterns of the sample and the constructed co-expression network for each stage, using a new method named single-sample edge perturbation network (SSEPN) for derivation of a co-expression network of a single sample. We will then infer the most significant edges (co-expressed gene pairs) in the network measured in terms of their level of impact in disrupting the overall network when altered, hence providing possible drug targets for the specific cancer case. A new algorithm will be developed to predict the impact of each such edge using information from the Connectivity Map, a popular database containing information about the impact levels of individual edges on an overall network when they are each perturbed, as well as information regarding the relationships between expression levels of individual genes and the survival rates of a cancer from the TCGA database. A few technical problems will be solved to accomplish this goal using a multi-variate regression and 0/1 integer programming. The deliverable of the project is an algorithm/program for predicting the most effective drug targets for a given cancer tissue sample and its transcriptomic data, along with a database for co-expression networks for each TNM stage of each cancer type in the TCGA database. We anticipate that this new capability should have wide applications for suggesting drug targets for individual cancer patients, assuming the relevant data are available. In addition, the methodology developed here should have applications in other areas of bioinformatics and systems biology.

个体化分子诊断和用药在癌症治疗上的作用日益突出，已成为癌症研究领域的重要前沿课题。相比于单个分子，网络生物标志物可在系统水平上更好地表征样本的状态。本项目利用基因表达数据计算基因间的共表达关系，在癌症各分期上分别构建精简的共表达网络；提出了一种构建单样本边扰动网络（SSEPN）的方法，通过计算网络扰动性对样本进行分期预测；进一步通过构建cox多元回归模型，获取与生存率最为相关的边扰动（即差异共表达基因），用于样本预后预测，并对其进行机理分析；充分利用CMap的数据，通过差异共表达分析获得药物对癌细胞共表达网络的影响，在蛋白质互作网络的映射下，结合癌症各分期网络及SSEPN的特征，通过0-1规划结合递归特征消除的方法，为个体提供最佳的用药方案。该研究成果对于癌症病人的个体化诊断和用药指导等问题具有重要意义，研究思路与方法也将为生物信息学和系统生物学等相关领域的深入研究起到积极的促进作用。

个体化分子诊断和用药在癌症治疗上的作用日益突出，已成为癌症研究领域的重要前沿课题。相比于单个分子，生物分子间网络可在系统水平上更好地表征样本的状态。 本项目以转录组和基因组数据为研究对象，以微环境压力下癌症演化的思想为指导，利用数据挖掘和复杂网络分析技术，识别个体癌症发生发展驱动力，从而实现精准诊疗。研究内容包括：构建单样本边扰动网络，在此基础上结合机器学习方法预测癌症样本分型；开发差异共表达分析方法识别癌症发展过程中一系列的生物网络变化、关键基因及生物过程；结合病人的基因型，综合评估药物对个体的逆转效应和副作用；进一步基于变分自编码器的因果推断方法，在去除混杂效应的同时进行反事实推断，识别个体致病驱动突变和评估个体药物效果，真正实现精准诊疗。.针对关键学术问题，围绕整体研究目标，项目顺利开展，项目申请人在国际著名期刊发表论文11篇，申请国家专利1项，发布了1个癌症精准药物推荐的平台。获得的重要结果包括 （1）提出了基于单样本边扰动网络的癌症精准分型方法；（2）从炎-癌转化角度,识别出驱动癌症发生的关键微环境驱动力，提出肿瘤发生发展的新机理模型；（3）提出了基于因果推断的个体致病驱动突变识别和个体药物推荐方法；.课题研究对于理解癌症发生发展原理，完善个体网络的基础理论和应用具有积极推动作用。该研究成果对于癌症病人的个体化诊断和用药指导等问题具有重要意义。

内容获取失败，请点击重试