Multiple QTL Mapping (MQM) provides a sensitive approach for
mapping quantititive trait loci (QTL) in experimental populations. MQM
adds higher statistical power compared to many other methods. The
theoretical framework of MQM was introduced and explored by
Ritsert Jansen, explained in the `Handbook of Statistical Genetics'
(see references), and used effectively in practical research, with the
commercial `mapqtl' software package. Here we present the first free
and open source implementation of MQM, with extra features like high
performance parallelization on multi-CPU computers, new plots and
significance testing.
MQM is an automatic three-stage procedure in which, in the first
stage, missing data is `augmented'. In other words, rather than guessing one
likely genotype, multiple
genotypes are modeled with their estimated
probabilities. In the second stage important markers are selected by
multiple regression and backward elimination. In the third stage a QTL is moved
along the chromosomes using these pre-selected markers as cofactors,
except for the markers in the window around the interval under study. QTL are
(interval) mapped using the most `informative' model through maximum
likelihood. A refined and automated procedure for cases with large
numbers of marker cofactors is included. The method internally
controls false discovery rates (FDR) and lets users test different QTL
models by elimination of non-significant cofactors.
R/qtl-MQM has the following advantages:
Higher power to detect linked as well as unlinked QTL, as long as the QTL explain a reasonable amount of variation
Protection against overfitting, because it fixes the residual variance from the full model. For this reason more parameters (cofactors) can be used compared to, for example, CIM
Prevention of ghost QTL (between two QTL in coupling phase)
Detection of negating QTL (QTL in repulsion phase)