遺傳模式未知下之家族資料抗變相關檢定

Abstract

以家庭資料為基礎探討疾病與標識基因之抗變相關性研究是近年來遺傳流行病學重要的研究方向之一，其優點在於此研究設計不受群體分層、非隨機交配與遺傳模式不確定的影響。一般家庭資料相關分析通常假設疾病之遺傳模式為累加模式，但實際上遺傳模式有可能為隱性、顯性或相乘等模式，當真實模式為隱性或顯性模式之下進行相關分析時，會發生檢定力下降的結果，採用抗變(意指可對抗遺傳模式的變動)相關檢定是解決上述問題的方法之一。在抗變遺傳相關研究領域中，直至目前為止僅有部分研究提出利用三元體資料與多病體核心家庭資料的抗變相關檢定，尚未有任何方法提出利用家族資料進行抗變相關分析。為了解決此懸宕問題，本研究採用羅吉斯回歸模式來建立疾病與候選基因間的相關聯結，並將截距項設定為家族效應參數，去描述不同家族間之異質性，在給定每個家族染病成員之下可建構家族資料之條件概似函數，再將回歸係數參數轉換為極座標後可以轉換後的新參數萃取出遺傳相關訊息，繼而推得計分統計量，再取最大統計量與小中取大有效抗變統計量進行疾病與候選基因之相關分析。模擬結果顯示此二種統計量皆為可用之相關檢定，並具有一定程度的檢定力，此足以印證本研究在遺傳模式未知的情況之下，確實提供了一個解決家族資料抗變相關分析的方法。

In recent years, development of robust association methods for analyses of familial data has become a crucial approach for identifying genetic variants that underlie human disease. The merit of such an approach can avoid the problem of population stratification, nonrandom mating and misspecification of the genetic model. It is known that the though commonly used genetic association test statistics have been shown applicable in practical studies, their performances rely on the mode of inheritance (MOI). As a matter of fact, these tests were derived under the additive model and they would lose testing power when the underlying genetic model is misspecified (e.g., as the true model is recessive or dominant model). Accordingly, development of robust association tests that have relatively stable powers over all plausible genetic models is necessary. Till now, robust methods have been widely proposed for case-parent trios and nuclear families, but not yet for pedigrees. In this dissertation, we use conditional likelihood, based on logistic regression model, to construct the association link between a disease and a marker for pedigree data. According to the likelihood function, the score statistic can be derived under a reparameterization procedure which is adopted to extract association information from the structure of regression coefficients. The maximum statistic and maximin efficiency robust test are derived for dealing with the unknown MOI problem in analysis of candidate-gene association. Simulation results indicate that the proposed test statistics are indeed robust against the effect of misspecification of MOI on testing powers.