摘要
In a typical composite interval mapping experiment, the probability of obtaining false QTL is likely to be at least an order of magnitude greater than the nominal experiment-wise Type I error rate, as set by permutation test. F2 mapping crosses were simulated with three different genetic maps. Each map contained ten QTL on either three, six or twelve linkage groups. QTL effects were additive only, and heritability was 50%. Each linkage group had 11 evenly-spaced (10 cM) markers. Selective genotyping was used. Simulated data were analyzed by composite interval mapping with the Zmapqtl program of QTL Cartographer. False positives were minimized by using the largest feasible number of markers to control genetic background effects. Bootstrapping was then used to recover mapping power lost to the large number of conditioning markers. Bootstrapping is shown to be a useful tool for QTL discovery, although it can also produce false positives. Quantitative bootstrap support—the proportion of bootstrap replicates in which a significant likelihood maximum occurred in a given marker interval—was positively correlated with the probability that the likelihood maxima revealed a true QTL. X-linked QTL were detected with much lower power than autosomal QTL. It is suggested that QTL mapping experiments should be supported by accompanying simulations that replicate the marker map, crossing design, sample size, and method of analysis used for the actual experiment.
In a typical composite interval mapping experiment, the probability of obtaining false QTL is likely to be at least an order of magnitude greater than the nominal experiment-wise Type I error rate, as set by permutation test. F2 mapping crosses were simulated with three different genetic maps. Each map contained ten QTL on either three, six or twelve linkage groups. QTL effects were additive only, and heritability was 50%. Each linkage group had 11 evenly-spaced (10 cM) markers. Selective genotyping was used. Simulated data were analyzed by composite interval mapping with the Zmapqtl program of QTL Cartographer. False positives were minimized by using the largest feasible number of markers to control genetic background effects. Bootstrapping was then used to recover mapping power lost to the large number of conditioning markers. Bootstrapping is shown to be a useful tool for QTL discovery, although it can also produce false positives. Quantitative bootstrap support—the proportion of bootstrap replicates in which a significant likelihood maximum occurred in a given marker interval—was positively correlated with the probability that the likelihood maxima revealed a true QTL. X-linked QTL were detected with much lower power than autosomal QTL. It is suggested that QTL mapping experiments should be supported by accompanying simulations that replicate the marker map, crossing design, sample size, and method of analysis used for the actual experiment.
