不同重組自交族群基因組結構模型下偵測兩個連鎖QTL的效力研究

Abstract

因爲重組自交族群(RI 族群) 比F2 族群或回交族群擁有更多的重組體以 及較多的同質基因型個體, 利於QTL 的偵測, 故偵測兩相鄰的連鎖數量性狀基因座(quantitative trait loci, QTL) 時, RI 族群提供較高的檢定力。 利用區間定位(interval mapping) 來探討兩相鄰QTL 的定位問題時, 其統計模式爲混合型常態模式, 爲了計算模式中的混合比例, 故必須知道RI 族群內三個基因座的基因型分佈。因爲馬可夫性質存在於F2 族群, 推導三個基因座的基因型分佈可由成對的兩個基因座的基因型分佈獲得。然而在 RI 族群內因爲多次的減數分裂喪失了馬可夫性質, 使三個基因座的基因型 分佈較難推算。Lynch and Walsh (1998) 和Jiang and Zeng (1997) 分別提出近似的方法, Kao and Zeng (2009) 則提出確切的方法來推算三個基因座的基因型分佈。本研究藉由這三種方法計算出模式中的混和比例, 並 比較它們偵測QTL 的檢定力差異。JZ 方法假設馬可夫性質於RI 族群中 依然存在, 故削弱了兩QTL 間連鎖的強度, 使檢定力比確切方法所得之檢 定力高。LW 方法假設在RI 族群中任兩個基因座的互換率可用重組體之 比例來代替, 使檢定力比確切方法低。本研究結果顯示近似的方法於RI 族 群中可能會高估或低估偵測QTL 的檢定力, 故在RI 族群中, 利用確切的 KZ 方法推算三個基因座的基因型分佈比較恰當。

The statistical model for QTL mapping is generally a normal mix- ture model and is usually proposed for F2 population. Recombinant inbred (RI) populations are also popular as they can provide greater power in mapping closely linked QTL by providing different genome structures to benefit QTL detection. In F2 population, the genome structure has the Markovian property. Therefore, the mixing propor- tions in the mixture model can be easily obtained by using pairwise genotypic distributions of two loci. In RI populations, their genome structures do not have such property due to multi-meiosis cycles. The derivation of the mixing proportions is more complicated as it in- volves the use of the genotypic distribution of three loci. We use three different methods, including Kao and Zeng’s method (Kao and Zeng, 2009), Lynch andWalsh’s method (Lynch andWalsh, 1998), and Jiang and Zeng’s method (Jiang and Zeng, 1997), to obtain the genotypic distributions of three loci for computing the mixing proportions and investigating the correlation structures between two putative QTL in the RI populations. Then, the powers of separating two linked QTL detection predicted by these methods are compared under the frame- work of regression interval mapping model (Haley and Knott, 1992). As compared to the power predicted by KZ method, it is found that JZ method, which has assumption of Markovian property, overesti- mated the power, and LW method, which replaces the recombination rate with the proportion of the recombinants in RI population, under- predicted the power. Numerical analyses are provided for illustration.