估計單核苷酸多型性遺傳力之新方法– 考慮變異間的相關性及變異與表現型的無母數關係

Abstract

遺傳力 (heritability)，普遍被用來衡量遺傳因子對表現型的變異之解釋能力。當全基因體研究僅有單核苷酸多型性 (SNPs) 的資料時，在這情況下所估計的遺傳力稱為單核苷酸多型性遺傳力 (SNP-heritability)。現有的估計單核苷酸多型性遺傳力之方法大多基於線性模型的架構；然而，基因結構在現實中可能是複雜的。在此，我們發展出新的估計單核苷酸多型性遺傳力之方法，其建立在廣義可加性模型 (generalized additive model, GAM) 下。此方法能夠同時考慮變異間的相關性，以及變異與表現型間之無母數關係。在配適廣義可加性模型前，考量到有些單核苷酸多型性會一起對表現型有所貢獻，我們會先將單核苷酸多型性進行分群；接著，針對每群中的單核苷酸多型性進行維度縮減，如此的處理可以保留單核苷酸多型性彼此間的相關性資訊。由模擬結果顯示，我們的方法在某些情況表現得比現存之方法 (如GREML、LDAK等)為佳。此外，除了總體的單核苷酸多型性遺傳力，群級單核苷酸多型性遺傳力亦同時被估計出來。我們的方法也可以應用到其他型態的DNA變異資料，也可以被延伸至其他類別型資料。

Heritability is a common measure that quantifies the proportion of phenotypic variation explained by genetic factors. When only single nucleotide polymorphisms (SNPs) are available in genome-wide analysis, the estimate of heritability is called SNP-heritability. Existing methods are mainly based upon the linear model structure; however, genetic architecture can be complicated in reality. Here we propose to consider the correlation between SNPs and the non-parametric relationship between variants and phenotype into the analysis based on the generalized additive model (GAM). Before fitting GAM, the analysis procedure begins with clustering SNPs and then dimensionality reduction is performed, for some SNPs may together contribute to the phenotype and this process can preserve the information. The results indicate that the proposed method performs better than current methods (e.g. GREML, LDAK) under some conditions. In addition, the cluster-level heritability can be estimated by our GAM-based method. This method can be applied to other types of DNA variation data, and can be extended to analyze different categorical data.