Interpretable Bayesian Non-linear statistical learning models for multi-omics data integration

项目摘要 最近的技术进步使得能够产生大量不同的多组学数据类型(例如， 基因组学、表观基因组学、蛋白质组学、转录组学)复杂疾病，如癌症、心血管疾病 和神经退行性疾病。整合来自这些异质疾病的多组学数据可能会有所帮助 在多个组学数据水平上解开潜在的生物学机制，在改善临床预测方面 结果，并改变医学，但同时也提出了重大挑战，以确定重要的 来自大量异质分子数据点(即数十万)的生物标记物。我们会 开发和应用新颖而强大的贝叶斯统计学习方法，以捕捉线性和非线性 多组学数据之间的关系。这些方法将被用来识别i)重要的预测途径和 其相应的重要分子；ii)具有临床意义的分子疾病亚型，以及iii)预测性 以及有助于组学数据之间的联合关联(或监管网络)的预后生物标记物 类型。建议的方法将应用于多个公开可用的数据集，如癌症基因组 以及从我们的合作者那里获得的非公开数据集。我们 将为我们的应用程序免费开发健壮、计算高效和用户友好的软件 方法：研究方法。

Project Summary Recent technological advances have enabled the production of vast amounts of diverse multi-omics data types (e.g., genomics, epigenomics, proteomics, transcriptomics) of complex diseases such as cancer, cardiovascular diseases and neurodegenerative disorders. The integration of multi-omics data from those heterogeneous diseases can help in unraveling the underlying biological mechanisms at multiple omics data levels, in improving prediction of clinical outcomes, and to transform medicine, but at the same time presents significant challenges to identify important biomarkers from a large size of heterogeneous molecular data points (i.e. hundreds of thousands). We will develop and apply novel and powerful Bayesian statistical learning methods that will capture linear and nonlinear relationships of multi-omics data. The methods will be used to identify i) important predictive pathways and their corresponding important molecules; ii) clinically meaningful molecular disease subtypes, and iii) predictive and prognostic biomarkers that contribute to the joint association (or regulatory networks) between omics data types. The proposed method will be applied to multiple publicly available datasets such as The Cancer Genome Atlas, dbGAP, and Genotype-Tissue Expression, and to non public data sets obtained from our collaborators. We will develop robust, computationally efficient, and user-friendly software free of charge for the application of our methods.