GENERALIZED LINEAR MIXED MODEL ALTERNATIVE PARAMETER ESTIMATION METHOD ON OIL PALM (ELAEIS GUINEENSIS) BREEDING

• Main Author: Sonhaji, Abdullah
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84288
• Institution: Institut Teknologi Bandung

Oil palm breeding methods are indispensable to increase production. Determining the parent that has a high potential to pass down superior traits to its offspring (progeny) is part of the most important step. The BLUP (Best Linear Unbiased Prediction) method is used for thet purpose. BLUP is a statistical method for estimating the random effect of the parent with the Linear Mixed Model (LMM). LMM is a special case of the Generalized Linear Mixed Model (GLMM). This research has two objectives. First, develop LMM on breeding data from two different locations. Second, develop an algorithm to estimate the parameters of the GLMM through Maximum Likelihood (LM) and use it to calculate BLUP. LMM assumes phenotypic data as a normally distributed response. This model is used in the case of experimental data with location design and kinship design. LMM is applied in two stages (TSLMM) with the aim of eliminating the effect of location and error while focusing on the calculation of the genetic effect inherited by the parents. The first stage of TSLMM is LMM considering location factors, and the second stage involves the parent random effect factor. This method works well when applied to actual field data. The phenotype tested is related to production, namely bunch weight, bunch number and its average weight. TSLMM is able to select the superior parents of the progeny. This model can use kinship information as a covariance matrix. This is useful if we are processing data from two experiments whose progeny sets do not overlap but are related. TSLMM has the advantage of simplifying the model. So it has a fast execution time and a high chance of convergence. The male and female parent varieties in the oil palm breeding data are very diverse. Likewise, the effect of interaction between parental genetic factors and their environment. This means that the number of random effect factors in GLMM will be very high. The total number of these random effect factors equal the integral degree of the likelihood function of GLMM. Consequently, it is difficult to find an analytical solution of this function. Therefore, an efficient integral approximation method is needed as a numerical solution. The Clenshaw-Curtis-Boyd (CCB) integral approximation is used in this case. The CCB has a user defined constant. This constant can be determined so that with a few points of Lobatto grid, even with a single point, the accuracy can be maintained. The CCB quadrature can be used in the GLMM parameter estimation algorithm (GLMM-CCB). Phenotypic random variable is not always normally distributed. Counting phenotypes, bunch number for example, can have Poisson distribution. The parent selection based on progeny phenotype is carried out with GLMM. GLMM-CCB with Poisson distribution and log link function is used to estimate BLUP. In this case, the random effect factors of the female and male parents. GLMM-CCB works well in determining parent that has superior traits. The selected parent is in accordance with its progeny performance.