Optimizing Gwas in Barley: Trade-Offs Between Statistical Power and Computational Efficiency Across Different Models

Genome-Wide Association Studies (GWAS) are pivotal for identifying quantitative trait nucleotides (QTNs) in crops like barley. However, the performance of different methods varies depending on population structure and computational demands. This study compared eight GWAS methods (GLM, MLM, CMLM, SUPER, MLMM, FarmCPU, BLINK, and ECMLM) using a simulated dataset from the World Barley Core Collection (WBCC) with 6,332 SNPs, 318 samples, a heritability of 0.7, and 15 QTNs across 30 replicates. Statistical power, false discovery rate (FDR), Type I error, and computational efficiency were evaluated at a 10 Mb window size through Power vs FDR/Type I error curves, Area Under the Curve (AUC) boxplots, QQ plots, Manhattan plots, and timing bar graphs. MLMM consistently outperformed other methods in balancing power and error control, followed closely by BLINK and FarmCPU, which also demonstrated high mapping resolution in Manhattan plots. GLM exhibited the highest false-positive rate, as seen in its QQ plot, while ECMLM underperformed despite theoretical advantages. GLM and MLMM were the fastest, whereas CMLM was the slowest, highlighting significant computational trade-offs. These findings suggest that MLMM is ideal for high-quality QTN discovery in barley, while BLINK offers a balanced approach for routine analyses. The study provides a framework for selecting GWAS methods in barley, emphasizing the importance of balancing power, error control, and computational efficiency in structured populations like the WBCC.