Accuracy of genomic prediction using group recordings

Group records are the alternative solutions for traits that are affected by group interaction and are difficult or costly to measure on individuals. We aimed to investigate how group size, group composition, and validation sets affect the prediction accuracy using group records. Group records were the average of individual records in a tank. We used the genomic selection method and three validation sets; across-family, next-generation, and within-family. Group records (with tank effect) of a subfamily (size=20) gave prediction accuracy of 0.54, 0.62, and 0.74 for the above order validation sets. Instead, data without tank effect raised the above accuracies to 0.60, 0.67 and 0.78. However, group records of a family (size=40) had 0.01-0.03 lower accuracy than the subfamilies scenario. The group records based accuracy further declined sharply by 0.17-0.12 when a group comprised two unrelated families. Overall, we got 0.10-0.40 reduced accuracy using group records instead of individual ones. Hence, grouping by families, a higher number of group records, data without tank effect, smaller group sizes, and a close relationship between validation and reference populations improved genomic prediction accuracy using group records.