基于多组学数据整合的药物响应预测及共同模式识别

Some drugs exert their function through binding to target proteins, some drugs target microRNA (miRNA), and some drugs intervene in the epigenetic control of gene expression. For the same cell lines, current technology has created omics data from different layers such as biological sequences, DNA methylation, gene expression, miRNA expression and protein abundance, and drug response data. This project will effectively integrate drug related multi-layer omics data. The research contents include: (1) based on the consideration that similar drugs exert therapeutic effects on similar cell lines, we will design many drug-drug, cell line-cell line similarity measures for drug response prediction by using many drug attributes and multi-layer omics data. (2) We will mine the common patterns existed between the multi-layer genomics and drug response data, identify drug associated cross-layer molecular regulatory interplay. (3) As to methodology, according to the features of the dataset such as high dimension, multi-layer and complex data structure, this project will design a new method for integrating these data, i.e. sparse multi-block singular value decomposition regression method, and use it to detect common patterns. The research of this project will lay the computational foundation of personalized medicine, where multi-omics data or molecular biomarker from individual patients can be used to predict drug response for themselves.

现阶段的药物有些通过结合靶蛋白来发挥其功效，有些以小分子非编码RNA（miRNA）作为治疗靶点，有些则干预基因表达的表观遗传控制。对于相同细胞系，现有技术产生了来自生物序列、DNA甲基化、基因表达、miRNA表达、蛋白质丰度等不同层次的组学以及药物响应数据。本项目将有效整合药物相关的多层次组学数据，研究内容包括：①基于“相似的药物治疗作用于相似的细胞系”考虑，利用多种药物属性和多层次组学数据，设计多种药物-药物、细胞系-细胞系相似性度量来预测药物响应。②挖掘多层次基因组学和药物响应数据之间的共同模式，识别药物关联的跨层次分子调控作用。③方法论方面，建立新的模型与算法，即稀疏多块奇异值分解回归方法，来整合这类高维、多层次、结构复杂的数据，并用于共同模式探测。通过本项目的研究，对于不同患者，可基于其多组学数据或分子标志物来预测药物响应，从计算角度为个体化医疗打下基础。

人类癌细胞系被用于研究癌症生物学和癌症治疗检验。最近已经有几个大的人类癌细胞系组，其包含上百个癌细胞系，并且具有基因组和药理学数据的特征。利用这些药物基因组学数据预测药物响应的能力可以促进精准癌症医疗的发展。虽然已经开发了多种方法来解决药物响应预测问题，但要获得准确的预测结果还存在许多挑战。基于相似细胞系和相似药物具有相似药物响应的事实，项目组采用相似性正则化矩阵分解（SRMF）方法，利用药物的化学结构和细胞系中的基线基因表达水平预测细胞系的抗癌药物响应。具体来说，药物的化学结构相似性和细胞系的基因表达谱相似性被考虑为正则化项，并被纳入药物响应矩阵分解模型。分析表明，所提出的数据整合方法能够提高细胞系中抗癌药物响应预测的准确性，并且可以识别与现有数据相一致的和新的药物-癌症基因关联，以及有助于药物重定位。项目组基于比较代谢组学分析了S-腺苷甲硫氨酸（SAM）对阿维菌素细胞内代谢物谱的影响机制，以有效提高阿维菌素产量。利用Petri网来建模DNA序列置信度值预测问题以及基因表达影响度预测问题。