The search for markers associated with a single major gene is fairly straightforward. Fit the linear model for each marker, and select the marker with the smallest significance level (p-value) or largest explained variation. The basic idea is to fit a linear model as in Section 1.1 using the genotype of each marker. The marker with the best fit is chosen as the QTL. One way to present results graphically is to plot a measure of fit against marker position in the genome. Markers that are closely linked should have similar fits. If more than one region of the genome shows strong association (small p-values), then it is wise to consider models involving multiple genes.
One of the main problems with this approach is sorting through volumes of printout from statistical packages! It is best to perform some sort of post-processing before printing to reduce material to the key results. Here is a summary for nine markers ordered in one linkage group. Marker WG6B10 was used in the previous section as a ``known'' QTL. Notice how the R-square and F Value numbers vary smoothly over the linkage group in the table below. The p values are all very small at one end of this group. It may be usefull to plot one or more of these against map position to identify potentially significant regions of the genome. In practice, certain regions of the genome tend to ``stand out'' as good candidates for putative QTL. However, as indicated in the previous section, it can be misleading to place too much weight on analysis based on only one QTL if two or more genes may be important.
Source R-Square F Value Pr > F WG2D11B 0.037 3.80 0.0540 WG1G8A 0.063 6.02 0.0161 WG3G11 0.090 9.14 0.0032 WG7F5A 0.114 12.09 0.0008 WG7F3A 0.189 21.42 0.0001 WG6B10 0.271 36.49 0.0001 WG8G1B 0.203 21.62 0.0001 WG5A5 0.248 31.33 0.0001 TG6A12A 0.248 28.35 0.0001 ACA1 0.295 40.22 0.0001
There is a danger in running many tests with the same data set, that eventually a significant result appears even if there is no QTL. This selection bias is probably not too serious, since the markers on the same linkage group are positively correlated [\protect\citeauthoryearOttOtt1991]. Some efforts have been made to adjust the critical value to account for this in different breeding systems [\protect\citeauthoryearLander and SchorkLander and Schork1994]. It is analogous to choosing a smaller significance level (say .005 instead of .05) for each test. One way to approximate this is to divide the overall significance level desired by the number of linkage groups.
A major shortcoming of this approach is that there is low power of detecting a QTL if it does not lie near a marker. Put another way, the difference between the genotype effects is partially hidden by recombination between the marker and the gene.