| Linear mixed models are the models in which both the fixed and the random effects contributes linearly to the response function. LMM have been widely applied in the quantitative genetics and breeding in livestock or crop. In this dissertation, we will extend the LMM to the three field, (1) Based on the theory of LMM, develop the advanced statistical approach to mapping QTL in the MAGIC populations, (2) Apply the LMM to partition heritability of quantitative traits across genomic annotations and (3) On the basis of linear models, apply the metabolomes to make the prediction for the agronomic traits.(1) Most standard quantitative trait locus (QTL) mapping procedures apply to populations derived from the cross of two parents. Compared to traditional QTL mapping populations, Multi-parent advanced generation inter-cross (MAGIC) populations are contructed from multiple parents via intercross and inbreeding for generations, which will have advantages in the detection resolution, statistical power and the inhibition of false positives. Besides, owing to the wide genetic basis, the results of QTL in the MAGIC population can be directly apllied to breeding. The greatest challenges of QTL mapping in MAGIC populations come from multiple founder alleles and control of the genetic background information. We developed a random model methodology by treating the founder effects of each locus as random effects following a normal distribution with a locus specific variance. We also fit a polygenic effect to the model to control the genetic background. To improve the statistical power for a scanned marker, we release the marker effect absorbed by the polygene back to the model. Both simulations and real data analyses demonstrated our proposed methods are able to effectively control false positives and improve the statistical power without increasing computational burden.(2) The researches in the other species have investigated what parts of the genomic annotations mainly contribute to the genetic variation undelying the complex traits. Here we extended the LMM to the five agronomic traits of 524 lines of rice to partition heritability according to genomic annotations. This project involved a total number of 3,616,597 SNPs, which were classified into four categories, regular, intron, exon and intergenic. Combining the summary of trait-associated SNPs (TASs) from genome-wide association studies (GWAS), we found that regular or intergenic region is the main source of the genetic variation of the five agronomic traits. We also implemented the different strategies of GBLUP to make the prediction for the traits through incorporating the various kinship matrix, that is, multiBLUP. We found that there were slight difference between different strategies of GBLUP.(3) Metabolites, as the products of the plant, could enhance our understanding of the molecular structure of the important agronomic traits for the rice and then are used as the predictors to help pick out the elite lines in the molecular breeding. Here we conducted the study of the 839 metabolites and the five agronomic traits, yield, heading date, plant height, grain length and grain width respectively, for the 524 lines. First we implemented the four models (LR, LMM, Bayes B, and LASSO) to investigate the association between metabolites and traits. Of the four models, LASSO performed best in controlling the false positive and power of detection. We also made the annotation of the metabolites detected by the LASSO. Finnaly, we made the prediction for the five agronomic traits using the metabolites based on the BLUP, Bayes B and LASSO models. In the most cases, except the LASSO method, using metabolites could gain more accurate predictability compared to GBLUP.The conclusions from the three studies have the potential to the application in the animal genetics and breedings, especially in the precision medicine. |
PDF Full Text Request