What proportion of declared QTL in plants are false?

Abstract

The false discovery rate (FDR) is the probability that a quantitative trait locus (QTL) is false, given that a QTL has been declared. A misconception in QTL mapping is that the FDR is equal to the comparison-wise significance level, alpha(C). The objective of this simulation study was to determine the FDR in an F(2) mapping population, given different numbers of QTL, population sizes, and trait heritabilities. Markers linked to QTL were detected by multiple regression of phenotype on marker genotype. Phenotypic selection and marker-based recurrent selection were compared. The FDR increased as alpha(C) increased. Notably, the FDR was often 10-30 times higher than the alpha(C) level used. Regardless of the number of QTL, heritability, or size of the genome, the FDR was < or =0.01 when alpha(C) was 0.0001. The FDR increased to 0.82 when alpha(C) was 0.05, heritability was low, and only one QTL controlled the trait. An alpha(C) of 0.05 led to a low FDR when many QTL (30 or 100) controlled the trait, but this lower FDR was accompanied by a diminished power to detect QTL. Larger mapping populations led to both lower a FDR and increased power. Relaxed significance levels of alpha(C)=0.1 or 0.2 led to the largest responses to marker-based recurrent selection, despite the high FDR. To prevent false QTL from confusing the literature and databases, a detected QTL should, in general, be reported as a QTL only if it was identified at a stringent significance level, e.g., alpha(C) congruent with 0.0001.