Empirical Comparisons of Different Statistical Models to Identify and Validate Kernel Row Number-Associated Variants from Structured Multiparent Mapping Populations of Maize

Although multiple statistical approaches for conducting GWAS are available, the power and false discovery rates of many approaches have been mostly tested using simulated data. Empirical comparisons of single variant (SV) and multi-variant maize (MV) GWAS approaches have not been conducted to test if a single approach or a combination of SV and Bayesian MV is effective, through identification and cross-validation of trait associated loci.

In this study, kernel row MPP number (KRN) data were collected from a set of 6,230 entries derived from the Nested Association Mapping (NAM) population and related populations. Three different types of GWAS analyses were performed: 1) single-variant (SV), 2) stepwise regression (STR) and 3) a Bayesian-based multi-variant (MV) models. Using SV, STR, and MV models, 257, 300, and 442 KRN-associated variants (KAVs) were identified in the initial GWAS analyses.