پردیس کشاورزی و منابع طبیعی دانشگاه تهران Iranian Journal of animal Science 2008-4773 49 1 2018 05 22 Impact of genotype imputation and different genomic architectures on the performance of random forest and threshold Bayes A methods for genomic prediction Impact of genotype imputation and different genomic architectures on the performance of random forest and threshold Bayes A methods for genomic prediction 145 157 67714 10.22059/ijas.2018.253649.653623 FA Yousef Naderi Assistant Professor of Genetics and Animal Breeding, Islamic Azad University, Astara Branch, Astara, Iran Journal Article 2018 03 06 Genomic selection using imputed genotypes can have an important role in increasing economic efficiency andthe genetic improvement of the threshold traits. The objective of this study was to: investigate the accuracy of imputation and to evaluate its effect on area under receiver operating characteristic (AUROC) of threshold BayesA (TBA) and random forest (RF) algorithms for discrete traits with different genomic architectures. Genomic data were simulated to reflect variations in heritability (0.30 and 0.10), number of QTL (108 and 1080) and linkage disequilibrium (low and high) for 27 chromosomes. To simulate a condition close to reality, we randomly masked markers with 50% and 90% missing rate for each scenario; afterwards, missing genotypes were imputed and imputation accuracy was estimated. In the last step, to evaluate the AUROC of TBA and RF, original or imputed genotypes were used. The accuracy of imputation was improved with increasing level of LD and decreased missing rate. The total average of AUROC values were 0.64 and 0.66 when using RF and TBA, respectively. Comparing to original genotypes, using imputed genotypes with 50% and 90% missing rate decreased the average AUROC about 0.013 and 0.02 for RF and 0.0018 and 0.026 for TBA, respectively. Despite the higher AUROC of TBA at different scenarios, RF showed a better performance in large number QTL. Generally, genomic prediction based on imputed genotypes (5K) can be implemented to reduce of the cost of a genomic evaluation. Genomic selection using imputed genotypes can have an important role in increasing economic efficiency andthe genetic improvement of the threshold traits. The objective of this study was to: investigate the accuracy of imputation and to evaluate its effect on area under receiver operating characteristic (AUROC) of threshold BayesA (TBA) and random forest (RF) algorithms for discrete traits with different genomic architectures. Genomic data were simulated to reflect variations in heritability (0.30 and 0.10), number of QTL (108 and 1080) and linkage disequilibrium (low and high) for 27 chromosomes. To simulate a condition close to reality, we randomly masked markers with 50% and 90% missing rate for each scenario; afterwards, missing genotypes were imputed and imputation accuracy was estimated. In the last step, to evaluate the AUROC of TBA and RF, original or imputed genotypes were used. The accuracy of imputation was improved with increasing level of LD and decreased missing rate. The total average of AUROC values were 0.64 and 0.66 when using RF and TBA, respectively. Comparing to original genotypes, using imputed genotypes with 50% and 90% missing rate decreased the average AUROC about 0.013 and 0.02 for RF and 0.0018 and 0.026 for TBA, respectively. Despite the higher AUROC of TBA at different scenarios, RF showed a better performance in large number QTL. Generally, genomic prediction based on imputed genotypes (5K) can be implemented to reduce of the cost of a genomic evaluation. https://ijas.ut.ac.ir/article_67714_1d4fcd53904e48d0976b197dd61c6e38.pdf