Unlocking complex co-expression network using graphical models

许多慢性病都很复杂，而且千差万别。它们可以受到多个基因的影响 结合生活方式和环境因素，患者可分为一种疾病 亚组，例如阿尔茨海默病(AD)的癌症亚型或分期。我们可以使用全基因组 基因表达数据来研究这些疾病的分子特征，这可能有助于理解 疾病病因和指导精准治疗。图形模型是估计复杂程度的强大工具 大量基因之间的网络相互作用。发展生物统计学和机器学习 使用基因表达数据来估计这种定向图形模型的方法是这项研究的主要目标 项目。为此，本项目包含以下研究活动：(1)给定已知的疾病亚型， 目标1开发了新的技术来联合估计多个无向/有向图形模型 每个子类型的型号。(2)在目标2中，我们考虑了没有预先定义疾病亚型的情况。 我们提出通过基因表达聚类来识别疾病亚型，聚类的不确定性是 纳入AIM 1中多个有向图形模型的估计。(3)最新的单细胞RNA测序 技术使研究人员能够对同一患者的多个细胞进行分析。目标3侧重于 估计多个有向图形模型(例如，针对肿瘤细胞的多个亚克隆，或多种类型的 脑细胞)使用一个患者的单细胞RNA-SEQ数据。所提出的图形模型的有效性 评估方法将使用癌症和AD数据分析进行演示。研究成果有 在理解和精确治疗这些疾病方面提供新的见解具有巨大的潜力。 此外，这些方法足够通用，可以应用于分析其他疾病的基因组数据，如 井。研究小组将分发计算效率高、用户友好的软件包， 研究出版物、学术报告以及与癌症研究和研究专家的合作 神经系统疾病。

Many chronic diseases are complex and very heterogeneous. They can be affected by multiple genes in combination with lifestyle and environmental factors, and patients of one disease can be divided into subgroups, e.g., cancer subtypes or stages of Alzheimer's disease (AD). One can use the genome-wide gene expression data to investigate these disease's molecular signatures, which may help understand disease etiology and guide precise treatments. Graphical models are powerful tools to estimate complex network interactions among a large number of genes. To develop biostatistical and machine learning methods to estimate such directed graphical models using gene expression data is the primary goal of this project. To this end, this project contains the following research activities: (1) Given known disease subtypes, Aim 1 develops novel techniques to jointly estimate multiple undirected/directed graphical models with one model per subtype. (2) In Aim 2, we consider the situation where disease subtypes are not defined a prior. We propose to identify disease subtypes by gene expression clustering, and the uncertainty of clustering is incorporated into the estimation of multiple directed graphical models in Aim 1. (3) Recent single cell RNAsequencing technology enables researchers to profile multiple cells from the same patient. Aim 3 focuses on estimating multiple directed graphical models (e.g., for multiple subclones of tumor cells, or multiple types of brain cells) using single cell RNA-seq data of one patient. The effectiveness of the proposed graphical model estimation methods will be demonstrated using cancer and AD data analysis. The research results have great potential to offer new insights on the understanding and precise treatments of these diseases. Furthermore, these methods are general enough to be applied to analyze omic data of other diseases as well. The research team will disseminate computational efficient and user-friendly software packages, research publications, academic presentations and collaborations with experts in cancer research and neurological diseases.