Estimation & Prediction

From High-Density SNP panel to High-Impact: Bridging Novel QTL Discovery and Genomic Prediction via FST Optimization

---

Abstract

Genomic selection (GS) significantly improves breeding accuracy compared to traditional methods but often plateaus at high marker densities. Prioritizing biologically influential SNPs using the fixation index (FST) can surpass this limit by identifying loci under selection pressure. While our previous studies demonstrated the utility of FST prioritization within known QTL regions, its application to de novo discovery when QTL architecture is unknown remains a critical challenge. This study evaluated a novel prioritization algorithm using simulated bovine data for a trait with a heritability of 0.4, controlled by 500 QTL. The population comprised 30,000 animals from the last generations (G9-G10) of a selection process, genotyped with a 600K SNP panel. To prioritize markers without prior knowledge of QTL positions, we developed a two-stage algorithm. Stage 1 (Signal Discovery and Raw Selection) utilizes a "Dual-Threshold" approach to identify candidate loci. Markers in the absolute top 1% of FST scores are retained unconditionally, while the top 10% "broad candidates" undergo physical isolation filtering-testing minimum separation distances of 5, 10, and 15 kb-to ensure distinct peaks. Genomic windows were evaluated using both 25 and 50 SNP margins defined around these peaks, qualifying only those exceeding the 25th percentile of genome-wide FST scores to establish the preliminary candidate list. Stage 2 (SNP List Optimization) refines this raw selection into fixed-size panels (20K, 40K, and 60K SNPs) by securing "Essential Peaks" (top 1% and filtered broad candidates) from Stage 1 and recruiting additional flanking markers using a proportional weighting system. This system allocates slots averaging normalized chromosome length and essential peak density, prioritizing higher FST SNPs for balanced genomic distribution. Genomic breeding values were estimated using GBLUP (BLUPF90+) to benchmark optimized lists against the full panel and standard prioritization strategies, selecting the top 1%, 5%, and 10% of SNPs based solely on FST scores. GS accuracy was evaluated using the correlation between true and estimated breeding values for 5,000 randomly selected animals from G10. Results indicated that while the full 600K panel achieved a baseline accuracy of 0.77 (estimated h2=0.37), the optimized panels reached a peak genomic accuracy of 0.76 (estimated h2=0.35). This peak performance was achieved in the 60K panels with 5 or 10 kb isolation distances when utilizing a 50-SNP margin. Increasing the SNP margin from 25 to 50 generally improved or maintained accuracy, capturing approximately 99% of the full panel's predictive potential. Crucially, the algorithm outperformed standard top 1%, 5%, and 10% FST strategies (accuracies: 0.60, 0.70, 0.73). By reaching a level of accuracy nearly equivalent to the high-density baseline with significantly fewer markers, this framework bridges biological discovery and genomic prediction when QTL architecture is unknown. Further investigation aims to refine the algorithm to fully capture the high-density baseline.

Keywords: 2026

How to Cite:

Toghiani, S., Aggrey, S. & Rekaya, R., (2026) “From High-Density SNP panel to High-Impact: Bridging Novel QTL Discovery and Genomic Prediction via FST Optimization”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2295878. doi: https://doi.org/10.31274/wcgalp.24380

Rights: 1