| Summary: | 碩士 === 國立臺北大學 === 統計學系 === 89 === This study uses the F2 population to investigate the differences between maximum-likelihood(ML) and regression(REG) interval mapping analytically and by simulation. In analytical investigation, we found that the two solutions set of ML and REG interval mapping methods have similar expressions but different contents. In the ML interval mapping, the conditional posterior expectations of QTL genotypes are used in estimation. In REG interval mapping, the conditional expectations of QTL genotypes are used in estimation. Therefore, the factors affecting the differences between the conditional posterior and conditional probabilities of QTL genotypes are the factors affecting the differences between ML and REG methods. The factors include heritability, the QTL position in an interval, interval size and the dominant effect. We find that the approximation of REG method to ML method becomes poor as (1) heritability becomes higher, (2) the QTL location is positioned toward the middle of interval, (3) the QTL is located in wider marker intervals, (4) the dominant effect becomes larger. Simulations confirm the analytical investigation . In simulation, the ML interval mapping tends to be more powerful and to give estimates with smaller MSE(mean square error) and large LRT(likelihood ratio test) in testing parameters. This implies that ML interval mapping can be more accurate, precise, and powerful in detecting QTL when compared to REG method.
|
|---|
