Table 2.

Genomic prediction accuracy in hybrid maize lines by prioritization of nonsynonymous SNPs in coding regions

				SNP prioritization
				Baseline	Observed PNC	Predicted PNC					SIFT conservation (1 − SIFT score)
Minimum percentile				None	None	0%	50%	90%	99%	99.9%	0%	50%	90%	99%
Validation	Ames-H ➔ NAM-H	Trait	DTS	0.775	0.773	0.776	0.777	0.775	0.773	0.771a	0.776a	0.777a	0.775	0.774
			PH	0.365	0.367	0.367	0.368	0.358	0.367	0.371	0.364	0.365	0.376	0.369
			GY	0.185	0.179	0.183	0.182	0.185	0.178	0.175	0.184	0.185	0.180	0.179
	NAM-H ➔ Ames-H		DTS	0.504	0.500	0.503	0.503	0.501	0.491	0.496	0.501	0.508	0.459a	0.454a
			PH	0.231	0.036	0.199	0.220	0.168	0.276a	0.111	0.182	0.137	0.079a	0.148
			GY	0.240	0.307a	0.254	0.267	0.296	0.330a	0.327a	0.243	0.240	0.187	0.200
	NAM-H (Leave-one-family-out)		DTS	0.404	0.405	0.402	0.401a	0.402	0.403	0.403	0.403	0.403	0.402	0.403
			PH	0.397	0.388	0.395	0.393	0.393	0.397	0.392	0.396	0.397	0.392	0.392
			GY	0.240	0.231	0.241	0.238	0.238	0.253a	0.249a	0.239	0.239	0.234	0.242

Genomic prediction models included effects of population structure variables (top three principal components from the Hapmap 3.2.1 reference panel in maize), genome-wide SNP effects, and prioritized effects of nonsynonymous SNPs. SNP prioritization consisted of weighting nonsynonymous SNPs uniformly (Baseline), or by a proxy for evolutionary constraint: observed phylogenetic nucleotide conservation (PNC), predicted PNC, or SIFT conservation (1 − SIFT score). For SNP prioritization by predicted PNC and SIFT conservation, SNP weights were also truncated to zero below their 0%, 50%, 90%, 99%, or 99.9% percentile. Validation: Ames-H ➔ NAM-H, training in a diverse hybrid panel (Ames-H) and validation in the Nested Association Mapping hybrid panel (NAM-H); NAM-H ➔ Ames-H, training in NAM-H and validation in Ames-H; NAM-H (leave-one-family-out), validation in each family in NAM-H after training in all other families. Trait: DTS days to silking, PH plant height, GY grain yield

^aPrediction accuracy was significant at the 5% level, based on random permutation of SNP weights. Underlined values indicate significant improvements over the baseline