Accuracy of Haplotype Reconstruction from Genotypes in Dairy Cattle

Haplotype inference is becoming increasingly important because of its applications in the study of diseases, population and evolutionary genetics, imputation, etc. Five popular methods for inferring haplotypes implemented in FIMPUTE, BEAGLE, SHAPEIT2, IMPUTE2 and FINDHAP were compared using 50K and 777k (HD) genotyping data and six different scenarios: no-parents, sire-progeny pairs, sire-dam-progeny trios, with or without pedigree information in a large Holstein cattle population. Methods were compared with respect to their haplotype reconstruction accuracy and computational efficiency. In the absence of parental genotypes and pedigree information, IMPUTE2 (with 2x default number of states) and BEAGLE were the most accurate methods, although they were only slightly more accurate than FIMPUTE. FIMPUTE generally outperformed other methods when genotype information of at least one parent was available. SHAPEIT2 showed intermediate accuracies and speed. FINDHAP was the fastest method, but generally the least accurate, especially for HD genotypes. FIMPUTE was 30-131, 87-786, and 353-1,400 times faster than BEAGLE, SHAPEIT2, and IMPUTE2, respectively. Therefore, FIMPUTE could efficiently handle (in 10-20 minutes) the datasets in this study (2,200-13,500 animals). The method implemented in FIMPUTE efficiently utilizes the information from close and distant relatives allowing accurate haplotype inferences, while maintaining computational efficiency needed for large datasets.