基于通路分析和深度学习算法的肿瘤靶向药物组学标志物及预测模型的研究

靶向药物主要通过识别特异性致癌基因的特征，阻断其表达的信号通路。不同靶向药物或用药过程耐药机制可能不同，而其共同点是相关基因突变或相关癌症通路被激活。本项目将结合BET抑制剂NHWD-870耐药的实际研究，提出通过提取癌症通路的统计特征获得耐药标志物和预测模型的方法。与传统的“差异基因表达”分析方法不同，本研究提出在利用先验知识和数据的基础上，从基因调控网络的拓扑结构和信息熵的变化上筛选“耐药”的标志物和通路，并重点研究如何使用深度自编码模型提取癌症细胞基因通路的特征，确定模型结构和参数，再按照迁移学习的思想和方法，支持新药临床试验使用标志物和预测模型进行亚组设计和分析。研究的主要内容：基于网络分析的耐药标志物（变量）筛选方法、基于KEGG通路的自编码深度学习模型的结构与算法、BET抑制剂NHWD-870耐药的主要标志物和通路分析、耐药预测模型、算法实现和血液监测代谢标志物的研究方法。

Targeted drugs block the signaling pathway of specific oncogenes by recognizing their characteristics. Drug resistance mechanisms may be different for different targeted drugs or medication process, however, the possible reasons attribute to the relevant gene mutation or activation of cancer pathways. In this project, we propose a novel statistical method to achieve drug resistance biomarkers and prediction model through the statistical characteristics of cancer pathways and illustrate this approach with real example NHWD-870 of BET inhibitor. Compared with the traditional “differential gene expression analysis” method, our project utilizes prior information and data to select the drug resistance-biomarker and pathways based on topological structure of gene regulatory networks and information entropy, particularly extracts the important characteristics of cancer-relevant gene pathway and determine the model structure and parameters, and further supports the subgroup design and analysis about identifying the biomarker and determining predictive models in clinical trial according to transfer learning idea. The main contents in our project include the development of drug-resistance gene selection methods based on network analysis, the establishment of novel autoencoder structure and algorithm based on KEGG pathway, the biomarker and pathway analysis of NHWD-870 resistance, the development of drug-resistance predictive model and its algorithm implementation, biomarker selection method by monitoring omics data.