A general method for the imputation of genomic data in crop species

点击翻译按钮获取中文摘要

The project will develop and test a toolkit to impute dense genomic information in crop breeding populations. Dense genomic information allows geneticists to unravel the genetics of traits using genome wide association studies, and breeders to speed up genetic improvement using genomic selection and genomics assisted breeding. These methods are most powerful when the density of genomic information is very high and the numbers of individuals genotyped are very large but the cost of collecting genotype information to build such datasets is prohibitive. A flexible and effective imputation toolkit will make it possible to build such datasets cheaply using imputed data. In a genetics and genomics context, imputation is the prediction of an unknown genotype in one individual from the known genotypes of other individuals (to give a trivial example, if individuals 'X' and 'Y' are known to have genotypes AA and CC respectively, then their offspring 'Z' is imputed to be AC). The value of imputation is that when combined with high-density genotype information from a few individuals, high-density information can be imputed for many individuals that have been genotyped at low-density, which vastly reduces the costs of datasets of dense genomic information.The project has three parts:-1. We will develop heuristic imputation algorithms that exploit the information in crop pedigrees, that correct pedigree errors and that generate approximate physical maps of the genome. Existing heuristic imputation algorithms, which were designed for livestock, do not work on crops because crop pedigrees are more complex than livestock pedigrees and crop data are of many different types, whereas livestock data is fairly homogeneous in type.2. We will develop probabilistic algorithms that integrate with the heuristic algorithms to produce a hybrid imputation algorithm for crops that combines the speed of heuristic algorithms with the flexibility and robustness of probabilistic algorithms. Existing probabilistic algorithms are too slow and require too much memory to work well with crop data.3. We will package the software apply it to a number of specific case datasets and breeding programs in KWS, which is one of the worlds four leading crop-breeding companies.