Building health intelligence with complex data on tumor cell states and therapy resistance

对癌症免疫治疗中涉及的分子生物学背景和免疫途径的全面描述是发展诊断和治疗的基础。我们在现代免疫检查点抑制剂的背景下进行研究，并努力建立和验证能够在患者亚群中检测长期生存益处的预测性标记。我们的生物采样和对临床数据的广泛注释使我们能够使用额外的协变量进行多变量分析，如突变负担、LDH水平和免疫渗透。在大量肿瘤和单细胞肿瘤分子异质性的预期图谱中，我们特别感兴趣的是黑色素瘤从脑、肺和皮肤转移的部位相关的分子差异。我们将进一步扩大我们的数据收集，通过注释表观遗传学和代谢学信息到已经注释的数据集，例如甲基化，人类白细胞抗原等位基因状态和毒性特征。这些分析得到了软件和易于使用的网络工具的开发，这些网络工具扩展了我们的PhenoTImE数据共享(翻译中心)。具体地说，我们的目标是用贝叶斯框架扩展Cox比例风险模型，以更好地预测多个队列中的事件间隔时间数据，因为人工神经网络在各种分析的黑色素瘤队列中还不能很好地预测在免疫检查点阻止转录的情况下的生存。为此，促进交换大型数据集是至关重要的。因此，我们支持将来自原始“组学”数据的量化与我们和研究单位其他项目生成的表型曲线合并。因此，我们通过必要的数据和元数据互联，帮助进行协作、及时的分析。我们还管理和维护我们在第一个供资阶段建立的计算服务器集群，用于分析对该联盟所有成员和合作者开放的数据集。通过这些努力，我们将我们的研究扩展到黑色素瘤以外的其他肿瘤实体，我们的管道最终将合并到更大的机构平台，如多组学肿瘤组织图谱和跨学科肿瘤委员会。

A comprehensive characterization of the molecular biological background and immunological pathways involved in the immunotherapy of cancer, in our case melanoma, are the basis for developing diagnostics and therapeutics. We perform research in the context of modern immune checkpoint inhibitors and strive to establish and validate predictive markers that can detect long-term survival benefit in a subpopulation of patients. Our biosampling and extensive annotation of clinical data allows us to perform multivariate analyses with additional co-variables such as mutational burden, LDH levels and immune infiltrates. Within the anticipated profiling of molecular tumor heterogeneity of bulk tumors and single cells we are especially interested in site-related molecular differences of melanoma metastases from brain, lung, and skin. We will further expand our data collection by annotating epigenetic and metabolomic information to already annotated datasets, with e.g. methylation, HLA allele status and toxicity profiles. These analyses are empowered by the development of software and easy-to-use webtools extending our PhenoTImE data share (‘translational hub’). Specifically, we aim at extending the Cox proportional hazards model with a bayesian framework to better predict time-to-event data across multiple cohorts as artificial neural networks do not yet perform well to predict survival under immune checkpoint blockade from transcriptomics in various analyzed melanoma cohorts. To this end, the facilitation of exchanging large datasets is crucial. Hence, we support the merging of quantifications from raw ‘omics’ data with phenotypic profiles generated by us and other projects in the research unit. Thereby we aid collaborative, timely analyses with the necessary interconnection of data and metadata. We also administrate and maintain the compute-server cluster that we established in the first funding phase for the analysis of datasets open to all members of the consortium and to collaborators. With these efforts, we extend our research to additional tumor entities beyond melanoma where our pipelines eventually will merge into larger institutional platforms such as multi-omics tumor tissue profiling and inter-disciplinary tumor boards.