通過連鎖不平衡評分回歸從全基因組關聯研究匯總統計中調整病例控制比失衡的統合分析

Abstract

全基因組關聯研究 (GWAS) 已成為遺傳生物標記物或與疾病之間風險評估的常用工具之一。在進行 GWAS 分析時，由於全基因組關聯研究個人資料(Individual data of GWAS)不足，在數據收集過程中通常以 GWAS 匯總統計量 (GWAS Summary Statistic) 作分析。連鎖不平衡評分迴歸 (Linkage Disequilibrium Score Regression, LDSR or LDSC)是透過尋找GWAS 匯總統計中的卡方統計量(Chi-square statistics)與連鎖不平衡評分(LD Score)的關係所建立的線性迴歸模型(Linear Regression Model)，它的斜率與遺傳力計算以及評估人群中的基因型與疾病之間的關係成正比，並透過使用截距反應種族混雜誤差。眾所周知，LDSR 是應用 GWAS 匯總統計進行遺傳力估計的最廣泛使用工具。但由於在使用LDSR時因GWAS 匯總統計量存在病例對照比率不平衡而引致偏差，而這種誤差尚未被研究。本研究透過使用模擬數據(Simulation data)以及應用真實GWAS匯總統計例子，分析病例對照比率不平衡對GWAS 匯總統計量以及對LDSR 的影響，並透過提出一個全新的簡單校正方法，從而降低GWAS匯總統計之間因對照比率誤差而引致的差異，並希望藉此提升使用LDSR的準確度。本研究亦初步展示了隨機病例對照比率的偏倚校正模型，並通過統合分析驗證了該方法的效果。

Genome-wide association studies (GWAS) have been one of the common statistical tools in genetic biomarker investigation or risk assessment between single nucleotide polymorphisms and phenotypes trait. When performing GWAS, GWAS summary statistics are commonly used during data collection due to insufficient of individual data. LD Score Regression (LDSR or LDSC) is a linear regression model based on the relationship between chi-square statistics from GWAS summary statistics and LD Score. Its slope can be used for estimation of heritability and reflect genotype-phenotype relationship in a population, while the intercept of LDSR reflects population stratification or other biases. It has been known that LDSR is the most widely used tool for heritability estimation using GWAS Summary statistics. However, there are some limitations such as case-control ratio bias that is not yet investigated. This research demonstrates the effect of case-control ratio imbalance bias in GWAS cohorts on LDSR using simulation cohorts and real GWAS summary statistics cohorts. Through proposing a novel statistical approach for reduction of case-control ratio bias in GWAS summary statistics, this study develop a preliminary model for bias correction with randomized case-control ratio cohorts and it is hoped that the accuracy can be improved when applying in LDSR. We also demonstrate a preliminary model for bias correction with randomized case-control ratio cohorts and examine this method with meta-analysis.