World Wide Web links to many, if not most, of the QTL mapping software packages can be found at a website created by Brian S. Yandell (Departments of Statistics and Horticulture, University of Wisconsin-Madison): www.stat.wisc.edu/biosci/linkage.html. Developers and users may wish to alert Brian (yandell@stat.wisc.edu) to packages which are not listed at the website. Other websites that maintain genetic software lists include the following:

• Evolution & Population Genetics Educational Database (EP-GED)
• Daniel Week's List of Documentation
• Wentian Li's Genetic Analysis Software List
• SoftSearch, GDB

The packages discussed at the workshop were:

• QTL Cartographer (Basten, Weir & Zeng, North Carolina State University),
• PLABQTL (Utz and Melchinger, University of Stuttgart-Hohenheim),
• Mapmaker/QTL (Lander et al., Whitehead Institute),
• Map Manager (Keith Manley et al.),
• QGene (Nelson and Tanksley, Cornell University).

• MapQtl by Johan van Oijen et al. (Centre for Plant Breeding and Reproduction Research (CPRO) of the Netherlands Agricultural Research Department (DLO)),
• the least squares interval mapping package developed by Chris Haley et al. (UK).

A concensus was reached that there is considerable overlap in the kinds of matings handled and statistics produced by the various QTL mapping software packages. Some packages are more sophisticated in their handling of genome search and multilocus QTL parameter estimation.

The software research and development needs identified by workshop participants are detailed below. Comments specific to certain packages or from software developers are indented. Further comments are most welcome.

1. Multilocus Interval Mapping. Several packages use interval mapping methods, but none have built in functions for multilocus interval mapping (simultaneous search of more than one chromosome segment).
   • MapQTL has an easy to use (and very fast) method to fit approximate multiple QTL models according to the MQM mapping procedure of R.C. Jansen (various papers, a.o. Jansen, 1993, Genetics 135: 205-211; Jansen, 1994, Genetics 138: 871-881). In the implemented part of MQM mapping, markers are used as cofactors to absorb the effects of nearby QTLs, thereby increasing the power for mapping other segregating QTLs. A future version will contain a procedure for automatic selection of cofactor markers; currently this has to be done by hand.
2. Discrete Traits. Although methods have been described for mapping discrete or ordinal traits, e.g., threshold traits and disease resistance traits with arbitrary scales, none of the packages explicitly handles such traits.
   • For non-normally distributed traits MapQTL contains the option to perform the nonparametric Kruskal-Wallis rank sum test per marker. Multi-locus models are not handled this way.
3. Dominant Markers. When populations are segregating for dominant and codominant markers, interval mapping functions are needed for dominant x dominant, dominant x codominant, and codominant x codominant marker brackets.
   • Most interval mapping packages only handle the latter. One package, QTL-Cartographer, handles mixtures of dominant and codominant markers.
   • MapQTL also handles all mixtures of dominant and codominant markers. MapQTL even handles mixtures of all segregation types in crosses with outbreeders, where per locus two up to four different alleles may segregate!
   • Dominant markers: Marker types (dominant are codominant) are freely choosable in PLABQTL.
4. Multivariate Analyses. Multivariate QTL mapping methods are needed to estimate QTL genetic correlations and correlated QTL effects and for estimating marker-assisted selection index parameters for multiple traits.
   • QTL Cargtographer has some facility for this (Jiang and Zeng, 1996, Genetics), although it is not well documented at this time.
5. Random and Mixed Effects Models. QTL or marker effects are often handled as fixed effects in QTL mapping experiments. There are circumstances where they might be handled as random effects, e.g., QTL or marker variance component estimates are needed for implementing marker-assisted index selection. There is merit to developing software for handling QTL or marker effects as random effects. Random effects interval mapping methods have been described, but have not yet been implemented in any QTL mapping software packages. QTL mapping software does not presently handle a full gamut of mixed linear model problems, e.g., experiment and environment designs, random polygenic variances,
   • Experiment and environment design problems are typically handled by estimating progeny least square means using Statistical Analysis System (SAS) software or other software for linear models and using the progeny means as input data for QTL mapping software.
   • PLABQTL computes an ad-hoc ANOVA to estimate varance components for QTL and QTLxenvironment. But still the procedure should be enhanced.
   • [Utz] miss[es] as a point the overestimation of genetic effects and the proportion of genetic variance explained by QTL, in other words the influence of model selection in the QTL detection process. Breeders should have unbiased estimates to evaluate MAS vs. conventional selection. Or is those hidden in point 5 ?
6. Marker-Assisted Selection Index Parameters. Software has not been developed for directly estimating marker-assisted index selection parameters (marker and genetic variances and marker index weights and scores). Estimates of these parameters must be patched together from the output of QTL mapping software or widely used statistical analysis packages (e.g., SAS). Some of the technical details for estimating MAS index parameters are not trivial and have not been worked out, e.g., selecting for two or more traits or across two or more generations. The ultimate package would permit the user to handle a wide variety of random and mixed effects estimation and selection index problems.
   • This looks for a daily application of MAS. Then, [Utz] thinks[], programs must be more user-friendly to allow people in the lab or field conducting MAS and combining observation and marker data. At the moment, programs are more for the skilled calibrator or biometrician.
7. Backcross Inbred Matings. Most of the packages surveyed handle backcross, doubled haploid, recombinant inbred, and F2, F3, ..., Ft progeny. None explicitly handle advanced backcross generations (BC2, BC3, ..., BCt) or inbred lines developed from backcross or advanced backcross generations (other than BC1S1).
   • These have been implemented in QTL Cartographer.
   • QGene may handle these matings as well.
   • MapQTL handles the following population types (one population is analyzed at a time): backcross (BC1), F2, any generation recombinant inbreds, (doubled) haploids and full-sib family of a cross pollinating species (i.e. a cross between non-inbreds).

• Christopher J. Basten, NC St U
• Johan van Ooijen, Centre for Plant Breeding and Reproduction Research (CPRO) of the Netherlands Agricultural Research Department (DLO)
• Kenneth F. Manly, Roswell Park Cancer Institute, Buffalo, NY
• H.F. Utz, U Hohenheim
• Zhao-Bang Zeng, NC St U