Clustering of the drug activities of the NCI-60 cancerous cell lines

对于大多数抗癌药物，对详细的作用机制知之甚少。即使已经确定了目标，如FDA批准的和临床试验中的药物，更广泛的脱靶效应仍然知之甚少。癌症是一种通过遗传和表观遗传改变而出现的疾病，这些改变扰乱了控制细胞生长、存活和分化的分子网络。为了开发更有针对性和更有效的癌症治疗方法，必须在这种网络化的系统级背景下了解和理解药物作用。我们结合了最先进的技术，将我们的大型药物化合物数据库（20，602种化合物）组织成一个在NCI-60癌细胞系中具有相似反应谱的化合物的连贯簇网络。由此产生的网络是高度一致的化合物类关系的现有的理解，分组已知的作用机制的药物成连贯的集群，连同新的化合物共享类似的响应曲线。与此同时，药物簇网络揭示了许多反应谱相关药物的簇，与富集已知作用机制药物的簇几乎没有关系或没有关系。这些药物化合物簇可能代表对新靶点和途径有活性的药物。为了表征这些潜在的靶向途径，我们将应用两种互补的分析方法，从每个集群的代表性的“枢纽”化合物。首先，将每种中枢化合物的活性特征与NCI-60细胞系相关的分子谱数据（转录本表达、基因拷贝数和基因序列变体）相关联，然后对具有显著相关分子谱的基因进行途径富集分析。其次，将应用弹性网络回归算法（一种机器学习方法），使用上述分子谱数据来学习药物反应的稳健的多因素预测因子。我们将建立这些方法的适用性，提出了几个'阳性对照'的结果，根据集群丰富的已知激酶抑制剂和DNA损伤药物。从这个基础上，我们将提出预测的目标途径和响应相关的基因集的几个完全未知的化合物集群。这些结果将使用基于化合物结构的方法来表征药物簇靶向特异性。通过对广泛研究的NCI-60癌细胞系的药物结构、活性和分子特征分析数据的整合，我们将能够将一个大型药物化合物数据库组织成功能相关的组，为进一步的重点研究提供基础资源。

For most anti-cancer drugs, relatively little is known about the detailed mechanism of action. Even where targets have been defined, as with FDA-approved and in-clinical-trial drugs, broader off-target effects remain poorly understood. Cancer is a disease that emerges though genetic and epigenetic alterations that perturb molecular networks controlling cell growth, survival, and differentiation. To develop more targeted and efficacious cancer treatments, it is essential to situate and understand drug actions in this networked, systems-level context. We have combined state-of-the-art techniques to organize our large drug compound database (20,602 compounds) into a network of coherent clusters of compounds sharing similar response profiles over the NCI-60 cancer cell lines. The resulting network is highly concordant with the existing understanding of compound class relationships, grouping known mechanism of action drugs into coherent clusters, together with novel compounds sharing similar response profiles. At the same time, the drug cluster network reveals numerous clusters of response profile-related drugs with little or no relation to clusters enriched for known mechanism of action drugs. These drug compound clusters may represent agents that are active against novel targets and pathways. To characterize these potential target pathways, we will applied two complementary analysis approaches to representative 'hub' compounds from each cluster. First, the activity profile of each hub compound will be correlated with molecular profiling data associated with the NCI-60 cell lines (transcript expression, gene copy number, and gene sequence variants), followed by a pathway enrichment analysis for genes with significantly correlated molecular profiles. Second, the elastic net regression algorithm (a machine learning approach) will be applied to learn robust, multifactorial predictors of drug response using the aforementioned molecular profiling data. We will establish the suitability of these approaches by presenting several 'positive control' results based on clusters enriched for known kinase inhibitors and DNA damaging drugs. From this foundation, we will present predicted target pathways and response-related gene sets for several entirely uncharacterized compound clusters. These results will be additionally focused using compound structure-based methods to characterize drug cluster target specificity. Integrating drug structure, activity and molecular profiling data over the widely studied NCI-60 cancer cell lines, we will be able to organize a large drug compound database into functionally related groups, providing a foundational resource for further, focused studies.