genehunter man page (version 1.1, dec 1996)


VARIANCE COMPONENTS

Summary:  run variance components analysis
No Arguments

This command looks for evidence of quantitative trait loci (QTLs). At each scan position, the program determines whether a significant amount of the variance in a quantitative trait can be attributed to a QTL at that position. Specifically, it calculates maximum likelihood values for the mean trait value (separately for each sex, if desired), additive and dominance variance components for the QTL, additive and dominance variance components for other, unlinked loci, and an environmental variance component. One or both of the dominance components can be optionally excluded. In addition, the program can incorporate covariate effects by estimating the regression of the trait value on a given covariate value. The significance of the QTL effects is tested by comparing the maximum likelihood model with another one in which the QTL variance components are constrained to equal zero. The likelihood ratio of the two models is used to calculate a LOD score which can be compared to a chi-squared distribution as in classical methods of QTL analysis.

Data preparation
Phenotype values should be included in the pedigree file, after the genotype values for each individual. Covariates should be listed immediately after the phenotypes. Multiple values may be entered, up to the numbers given by the constants MAX_PHENOTYPES and MAX_COVARIATES in npl.h. Each phenotype should be indicated in the map file by a single line reading "0 2" followed by five empty lines (These are needed to maintain consistency with the LINKAGE file format. The data expected by LINKAGE in these lines are not used by Genehunter and can be excluded). Each covariate should be indicated with a single line reading "4 0". In addition, the total number of loci (the first number on the first line of the map file) should include the number of phenotypes and covariates, as well as the number of markers and qualitative traits.

Running the program
When the "variance components" command is entered, the user is prompted for names for the output files and is then asked whether to include dominance variance components for the unlinked loci and for the QTL. The user is then given the option of entering starting estimates for the model's parameters, rather than letting the program come up with it's own estimates. This option is provided because the program's ability to converge on the maximum likelihood values is sometimes sensitive to the starting guesses. Trying out different starting values and seeing whether the same result is obtained provides a check on the correctness of the results. This should probably be done with all analyses, but is especially needed if the program is yielding odd results, such as negative or unrealistically high LOD scores. If manual input is chosen, the program first displays the total variance of the trait value being examined, as well as the mean trait value (separately for males and females if this option has been chosen). These figures can be helpful in choosing starting values.

Output
The output file shows a LOD score for each scan position, along with estimates of the means, variance components, and covariate regression coefficients for that position. The corresponding estimates for the null model are also reported. Because the program can sometimes fail to converge on an estimate for some positions, the output indicates for each position whether convergence occurred. When it does not occur, the output shows the estimates for the last position which did converge. If the program frequently fails to converge, it may be necessary to raise the maximum number of iterations allowed in the estimating algorithm (MAXITS in the file varcom.c). If postscript output is switched on, the program also produces two graphics files. One contains a plot of LOD score versus position, and the other a plot of the proportion of total phenotypic variance accounted for by each component of the maximum likelihood model, versus position.