We have seen how to carry out a genome scan using interval mapping, assuming either a normal model or using a more robust procedure, and how to test the significance of peaks in the Z or LOD plot. What is left to do? Unfortunately, quite a lot. As explained at the beginning of the week, most QT differences (and the same probably applies to qualitative traits) will require differences at a number of loci to explain them, and interval mapping is clearly a one locus at a time approach. A modification has been given to permit the inclusion of effects due to differences at specified loci, while searching for others, but this is still far from a general attack on the problem. The truth is that there are not, as yet any satisfactory multilocus scanning procedures in use.
Many ideas exist, and here is one that was explored by Karl Broman in his
Berkeley PhD thesis. Firstly, forget interval mapping. It requires too much
computation to be feasible for 3, 4 or more loci. And besides, as Karl and
others have shown, its advantages in locating QTL with current marker densities
and sample sizes, are small, in comparison with methods focussing solely on the
analysis at the markers. Forget testing, because finding QTL is not a testing
problem: the loci are there, and the problem is finding them, and their mode
of action. Karl viewed the problem as one of model selection, and the task as
finding ways of searching through the huge number of multilocus models, and
comparing them in their ability to explain the associations between marker data
and phenotype. Work along these lines has barely begun, and it is far from
clear whether it is likely to be successful. In my view, much more
attention needs to be paid to information in data concerning the QT in
the parental lines and the 
progeny, as we seek clues to reduce
the number of possible models to a realistic number through which we
can search.