Using whole-genome sequence (WGS) data to identify the causative variants and improve genomic prediction is of current research interest. However, single nucleotide polymorphisms (SNP) chips are still the primary source for genomic predictions. Regular SNP chips only include a small number of SNP. Therefore, more accurate genomic predictions would be expected with WGS data. The objective of first study was to investigate the impact of using preselected variants from WGS for large-scale single-step GBLUP (ssGBLUP) genomic predictions in maternal and terminal pig lines separately. Genomic predictions with regular SNP chip data were compared with preselected SNP sets. Preselection of SNP relied on genome-wide association studies (GWAS) and linkage disequilibrium (LD) pruning. A second study aimed to explore the use of selected WGS variants in a multi-line ssGBLUP genomic evaluation (MLE), which comprised over 200,000 sequenced/imputed animals. A multi-line GWAS was conducted to preselect WGS variants, and unknown parent groups (UPGs) or metafounders (MFs) accounted for genetic differences among lines in a joint evaluation. Those first two studies reported small to no gain in accuracy of genomic prediction with WGS data. To explore the possible reasons for the limited gain in accuracy of genomic prediction with WGS data, a simulation study with different effective population sizes (Ne) was carried out in the third study. We investigated different discovery set sizes in GWAS, relating them to the limited dimensionality of genomic information. The selected variants based on different GWAS sample sizes were then added to simulated SNP panels that mimicked regular chips used commercially. Populations with smaller effective sizes (Ne = 20) require more data to capture causative variants, whereas for large populations (Ne = 200), using the number of genotyped animals equal to that of the largest eigenvalues explaining 98% of the variance of the genomic relationship matrix suffices. However, only a small proportion of the causative variants can be discovered if those genotyped animals do not have many progeny records. Even when several causative variants are preselected, their impact on ssGBLUP genomic predictions is minimal because medium-density commercial SNP chips already account for most of the information added.