Genomic prediction of unordered categorical traits: an application to subpopulation assignment in German Warmblood horses

Table 2 Measures of relationship and separability between subpopulations

Base	A	G	\({F_{st}}\)	\({h_{Ped}^2}\)
12	0.0002 (0.0075)	−0.0290 (0.0321)	0.0151 (0.0196)	0.9809 (0.0158)
13	0.0029 (0.0269)	0.0085 (0.0447)	0.0029 (0.0040)	0.9179 (0.0675)
14	0.0008 (0.0143)	0.0123 (0.0334)	0.0070 (0.0100)	0.8444 (0.0779)
23	0.0003 (0.0084)	−0.0176 (0.0377)	0.0109 (0.0147)	0.9442 (0.0269)
24	0.0000 (0.0000)	−0.0269 (0.0252)	0.0117 (0.0176)	0.9207 (0.0700)
34	0.0001 (0.0057)	0.0003 (0.0262)	0.0097 (0.0137)	0.7967 (0.1376)

Base	\({h_{Marker}^2}\)	pEV	nEV	CV-Accuracy
12	0.9575 (0.0180)	0.0133 (0.0056)	8.6981 (3.6581)	0.9832
13	0.8828 (0.0546)	0.0480 (0.0227)	25.1528 (11.8885)	0.7581
14	0.9233 (0.0417)	0.0122 (0.0071)	4.2579 (2.4701)	0.9457
23	0.9441 (0.0292)	0.0209 (0.0086)	11.8532 (4.8888)	0.9418
24	0.9214 (0.0532)	0.0061 (0.0043)	2.3697 (1.6806)	0.9974
34	0.9253 (0.0556)	0.0176 (0.0100)	4.6038 (2.6294)	0.9544

The averages (standard deviations) of pair-wise additive genetic relationship (A), genomic relationship (G), average \(F_\text {ST}\) values of all markers, pedigree-based heritability on the liability scale (\(h_{Ped}^2\)), marker-based heritability on the liability scale (\(h_{Marker}^2\)) for the trait “subpopulation assignment”, percentage (pEV) and number (nEV) of eigenvectors included in the variable selection threshold model and the prediction accuracy of subpopulation assignment (CV-Accuracy) of a leave-one-out cross-validation using SVM. Base indicates the classes used in the pairwise contrasts. Numeric indicators for subpopulations are: 1 Hanoverian, 2 Holsteiner, 3 Oldenburger, 4 Trakehner

ISSN: 1297-9686