复杂疾病基因与环境互作模型分析方法研究

Complex diseases root in genetics, but are influenced by environmental factors. The degree of influcence largely depends on how sensitive genetic factors respond to environmental stimuli. Currently many gene-environment (G×E) interaction methods are based on linear models. However, when genetic responses to environment stimuli do not follow a linear relationship, this could leads to large false negatives. In this investigation, we will relax the linear assumption and propose a nonlinear G×E interaction model under the statistical varying-coefficient model framework which overcome problems many methods are facing such as strong assumption on linear interaction as well as interpretability issue. Following nonparametric theory, we will provide effcient estimation and inference procedure. Built upon gene-set or pathway-based association studies with high-dimensional SNP variants, we will propose a penalized high-dimensional variable selection framework to select potential genetic players in a gene set or pathway that are sensitive to environment changes. The model will be extensively tested with large scale simulations as well as applications to three real data sets such as Type 2 diabetes. The developed models fit better to biological realm, and have particular power to attack long-standing genetic questions regarding the interplay between genes and environments. User friendly computational software will be made available. The success will provide important tools to facilitate the process of disease gene identification, and advance the discovery of novel genes and pathways to facilitate identification of drug targets to enhance public health.

复杂疾病源于遗传变异，但其发展却受环境的影响与调控，并取决于遗传变异对环境改变的敏感程度。目前基因与环境的交互作用多用一般线性模型来预测，由于基因对环境刺激的反应不一定都遵循线性变化，线性模型的应用有一定的局限性，且易致假阴性。本项目提出非线性基因与环境互作模型，基于群体关联分析原理，将变系数模型拓展到基因-环境互作研究，以解决目前基因-环境互作方法中对模型假设性过强、及可解释性和可预测性偏离实际问题等不足。项目拟通过非参数理论，模拟证实变系数模型原理，完成便于推广应用的相关软件；针对基因组或疾病通路内高维SNP数据，提出基于惩罚估计理论的高维互作处理技术，使结论更稳健于互作的生物学原理；并通过2型糖尿病等三份SNP数据和大型计算模拟，拟合非线性变系数模型，对其有效性和实用性进行评价，为复杂疾病的发病机制研究、基因诊断和预防提供研究工具与方法，在统计技术和实际应用上获得双赢。

复杂疾病源于遗传变异，但其发展却受环境的影响与调控，并取决于遗传变异对环境改变的敏感程度。目前基因与环境的交互作用多用一般线性模型来预测，由于基因对环境刺激的反应不一定都遵循线性变化，线性模型的应用有一定的局限性，且易致假阴性。本项目提出非线性基因与环境互作模型，基于群体关联分析原理，将变系数模型拓展到基因-环境互作研究，以解决目前基因-环境互作方法中对模型假设性过强、可解释性和可预测性偏离实际问题等的不足。基于非参数建模理论，我们提出了以下的模型和开展了以下的工作：.1. 就基因与环境互作的非线性假设，基于二分类疾病表型，构建了非线性的基因与环境互作模型；同时构建了以基因为单位的基因与环境互作非线性模型。传统的基因与环境互作模型考虑的是单个SNP位点与环境的互作，此模型考虑以基因为单位，从而真正地实现了以基因为单位的基因与环境互作；.2. 开发了基因与基因与环境的三重互作模型。采用此方法我们可以研究环境因子是如何调控基因与基因互作从而影响疾病的发生；.3. 构建了基因与多个环境因子整体互作的统计模型：此模型可以研究多个环境因子是如何作为一个整体与基因互作而影响疾病的发生发展；.4. 针对基因组或疾病通路内高维SNP 数据，提出基于惩罚估计理论的高维互作处理技术，使结论更稳健于互作的生物学原理；.5. 通过大量的模拟证实模型的小样本拟合效应，对其有效性和实用性进行评价, 相关计算程序采用R语言编写，可以广泛推广；.6. 应用于2 型糖尿病和出生体重等SNP数据，通过实际数据分析来验证模型的实证性；.我们开发的方法和程序为复杂疾病的发病机制研究、基因诊断和预防提供了研究工具与有效的方法。