A review of deep learning applications for genomic selection

BMC Genomics

Table 5 Prediction performance in terms of Pearson’s correlation for the simulated and real Arabidopsis datasets (Pook et al., 2020)

A). Predictive ability on different traits with
Trait architecture	GBLUP	BayesA	EGBLUP	MPL	CNN	LCNN
10 additive QTL	0.639	0.66	0.635	0.637	0.627	0.666
1000 additive QTL	0.516	0.538	0.543	0.524	0.538	0.606
10 epistatic QTL	0.511	0.527	0.519	0.503	0.491	0.572
1000 epistatic QTL	0.416	0.414	0.448	0.395	0.403	0.401
10 locally linked epistatic QTL	0.488	0.501	0.529	0.504	0.544	0.625
1000 locally linked epistatic QTL	0.524	0.523	0.541	0.519	0.517	0.51
B). Predictive ability for the Arabidopsis traits
Trait architecture	GBLUP	BayesA	EGBLUP	MLP	CNN	LCNN
Average predictive ability (all)	0.39	0.382	0.382	0.316	0.312	0.34
Average predictive ability (training set < 100)	0.404	0.39	0.399	0.3	0.299	0.326
Average predictive ability (100 < training set < 250)	0.364	0.358	0.354	0.318	0.311	0.327
Average predictive ability (training set > 250)	0.477	0.477	0.472	0.358	0.37	0.456

ISSN: 1471-2164