Skip to main content
. 2012 Nov 8;13:100. doi: 10.1186/1471-2156-13-100

Table 4.

Bias* of genomic predictions from different methods, obtained for five traits of a mice population

Splitting Method Trait
W6W WGS BL %CD8+ CD4+/CD8+
Within
POL
−4%
1%
−1%
1%
1%
emBayesB
−62%
24%
2%
−42%
−52%
RR_GBLUP
−9%
8%
13%
2%
4%
SS_BY
−18%
22%
-
5%
5%
SS_ABS
−35%
4%
−30%
5%
17%
RKHS
−1%
1%
−1%
1%
2%
SVR
−4%
3%
−6%
5%
4%
BayesCpi
−26%
66%
81%
−568%
−91%
BayesC
−43%
263%
69%
−397%
−420%
LASSO
−85%
53%
2%
−73%
−73%
RF
−1%
1%
1%
0%
1%
Across POL
-
-
-
-
-
emBayesB
−51%
5%
3%
−58%
−87%
RR_GBLUP
−31%
8%
8%
−2%
14%
SS_BY
−44%
18%
-
−1%
17%
SS_ABS
−54%
−7%
−38%
−16%
31%
RKHS
−23%
4%
−4%
9%
10%
SVR
−24%
6%
−8%
14%
10%
BayesCpi
−15%
83%
46%
−719%
−196%
BayesC
23%
280%
115%
−239%
−359%
LASSO
−116%
40%
7%
−70%
−98%
RF −13% 2% −1% 5% 5%

*Average of ten replicates. Bias was measured as the average difference between observed and predicted phenotypes of testing set and is presented as a proportion of the standard deviation of each trait (in %). Trait = weight at 6 weeks (W6W), weight growth slope (WGS), body length (BL), percentage of CD8+ cells (%CD8+), ratio between CD4+ and CD8+ cells (CD4+/CD8+). Splitting= splitting strategy in cross-validation (within or across-family).