SRmapqtl uses the technique of stepwise regression to search for QTLs. For forward and backward regression, it simply ranks the markers for their effect on the quantitative trait. In forward stepwise regression (FS), each marker in turn is tested for its effect on the quantitative trait using linear regression. That marker with the largest partial F-statistic is assigned rank 1 and included in all subsequent analyses. Step two tests all the remaining markers, and assigns rank 2 to the marker with the largest partial F-statistic. This is repeated until all the markers have been ranked.
Backward elimination regression (BE) starts with all markers in the model. In the first step, each marker in turn is removed and a partial F-statistic is calculated. That marker with the smallest partial F statistic is given the lowest rank and removed from subsequent analyses. This is repeated until all the markers have been ranked.
The above methods seek only to rank the markers: They make no effort to determine whether adding or deleting a marker makes a significant difference for the fit of the model to the data. A third method (FB) is to start with forward stepwise regression, but only keep adding markers while the p-value of the partial F statistic of the marker to be added is below a defined threshold, . When a step is reached in which no more markers can be added, all of the markers are retested to see if they are still significant. Each marker in turn is deleted from the model, a p-value is calculated for the partial F-statistic, and if the p-value is greater than a specified level , it is deleted.
You can put a hard limit on the number of steps in the regression analysis with the -u option. The program has internal limits based on the sample size (it won't allow more parameters than sample points), but you can lower this further. It defaults to 100, which is generally more markers than you will need to rank. If SRmapqtl seems to run forever, you might run it with this parameter set to a reasonable number, say 20.
As with LRmapqtl, any otraits that begin with a plus sign are also used in the regression model. Unlike LRmapqtl, no interaction terms are used. The command line parameters for SRmapqtl are listed in Table 3.3. One added feature is that if you use the -t option with an integer value one greater than the number of traits, then all traits will be analyzed in turn.