Multi-omics analysis

Using CADD scores to improve genomic prediction accuracy for production and reproduction traits in Piétrain pigs using whole genome sequence data

---

Abstract

In current practice of animal breeding, genomic best linear unbiased prediction (GBLUP) remains the primary standard for genomic prediction (GP). However, this model assumes that single nucleotide polymorphisms (SNPs) contribute equally to genetic variation, which does not reflect the reality that SNPs closely linked to causal variants explain more variation than others. Incorporating biological information can refine GP models by predicting SNP impact. For example, one could use combined annotation dependent depletion (CADD) scores, which integrate various genomic annotations to reflect the likelihood of variants being deleterious. Because CADD scores capture general importance of a variant across the genome, a high CADD score indicates a high potential impact of a SNP on phenotype, regardless of the specific trait. Our objective was to determine the benefit of using CADD scores to select SNPs from whole genome sequencing (WGS) data in pigs. We analyzed 4,627 Piétrain pigs for three production traits (muscle depth, average backfat, daily gain) and two reproduction traits (number of born piglets, average birth weight litter). We assigned CADD scores to SNPs by two strategies: (1) using the CADD score at the position of the SNP (CADD-SNP), and (2) using the maximum CADD score of all SNPs within a fixed-size window centred around each SNP (CADD-window), with the window size defined based on linkage disequilibrium. Selection thresholds were considered between top 1% and 10% with intervals of 1%, and between top 10% and 90% with intervals of 10%, based on CADD scores. For each thresholds, we randomly selected the same number of SNPs for comparison. We used a GBLUP model for GP and compared prediction accuracies of all SNP subsets with those obtained using the full WGS data after quality control. Compared to the full WGS data, selecting the top 10% of SNPs based on CADD-SNP scores performed best across all thresholds, slightly improving prediction accuracy for four out of five traits, but not for muscle depth. For most traits, accuracy was lower at the top 1% than that obtained using the full WGS dataset, became higher at the top 6%, peaked at the top 10% with an average improvement of 0.001, and then gradually decreased to levels comparable to those using the full WGS dataset. For CADD-window scores, accuracy decreased across all these thresholds. This decrease was largest at 1%, with average difference of 0.214 across the five traits. Random selection had little impact on prediction accuracy. In conclusion, using CADD-SNP scores to select SNPs improved prediction accuracy using WGS data, but the improvement was limited and the benefit depended on the trait of interest. This differs from our previous study, which found that using CADD-window scores to select SNPs improved prediction accuracy using SNP chip data.

Keywords: 2026

How to Cite:

Fu, C., Calus, M., Bink, M. & Duenk, P., (2026) “Using CADD scores to improve genomic prediction accuracy for production and reproduction traits in Piétrain pigs using whole genome sequence data”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2284789. doi: https://doi.org/10.31274/wcgalp.23617

Rights: 1