12.2 Sample multivar parameter file

The pedigree file for PolyEM programs is similar to `ped73.ped', which was used in most of the previous examples.

The first three entries in each line consist of the individual's name, father's name and mother's name. Integers starting with the fourth column (usually gender) can be fixed effects (gender, age class, etc.) or discrete phenotypes.

For quantitative traits, real numbers follow the names and integers. These real numbers represent trait measurements. Missing values are coded with integer part `999', such as 999.5 in the following example.

Here is part of the pedigree file `polyem.ped'. This file can be found in the `PolyEM' subdirectory of `MORGAN_Examples'.

input pedigree size 90 input pedigree record names 3 integers 3 reals 2 **************************************** 1 0 0 1 1 0 0.0246 -1.0125 2 0 0 2 1 0 -0.5978 1.5963 3 0 0 1 1 0 -0.8124 0.5662 4 0 0 2 1 0 0.4334 1.7721 5 1 2 1 1 0 0.1802 -1.4672 6 1 2 1 1 0 -1.7557 0.8091 7 3 4 2 1 0 999.5 999.5 8 3 4 2 1 0 1.9128 0.9780 9 0 0 2 1 0 0.9530 2.3473 ...

Below is the example multivar parameter file, `polyem.par'.

input pedigree file `polyem.ped' select trait 1 select trait 2 effects 1 2 start residual covariance -0.09 start additive covariance -0.0017 start residual variance 1.10 0.65 start additive variance 0.037 0.0288 fit residual covariance 1 fit additive covariance 1 fit environmental model output spacing 20 EM iterations limit EM iterations 200

multivar can fit a polygenic model with one to five traits, which can be modeled as dependent and/or independent. A `select trait' statement is required for each trait to be modeled (up to 5 traits). This statement indicates which column of real numbers the program is to examine. In the example, the statement `select trait 1' indicates that the first column of real numbers is to be examined. The statement `select trait 2 effects 1 2' indicates that the second column of real numbers is to be examined as well. The additional `effects 1 2' portion of the statement is optional and indicates that two fixed effects (also called covariates) are to be modeled for trait 2. The integers give the location of the fixed effects (covariates) starting with column 4 in the pedigree file. In this example, the fixed effects are to be found in columns 4 and 5. Important: a fixed effect location of `1' indicates that the effect value will be found in column 4. The most commonly modeled fixed effect is gender, which, if present, resides in column 4 of a MORGAN pedigree file.

The statement `start trait mean' allows the user to specify the starting trait mean for a selected trait. Since no `start trait mean' statement is included after either of the `select trait' statements, both of the initial means are computed by the PolyEm program. Similarly, one may specify the initial values for each effect with the statement `start trait I effect M X1 X2 ...', where `I' is the trait number, `M' is the effect number and `Xi' is the starting value of the ith level (i = 1, 2, 3, ...). These starting values represent deviations from the global mean. The starting values are normalized so that their weighted sum is zero (weighted by the number of individuals in that level). When using the `start trait I effect M X1 X2 ...' it is important to know that if more levels are present in the column of numbers corresponding to the trait `I' in the pedigree file than are specified in the `start trait' statement, PolyEm programs will compute the starting value(s) for these additional levels. Since the program will not issue a warning or error message, it is important to always check the output to confirm that the number of levels present in the file was as intended. Since the `start trait I effect M X1 X2 ...' statement is not included in this example, thePolyEm program will compute the initial values of the effect.

Initial values for additive and residual variances and covariances are specified in the next four statements. These statements are required. With the variance statements, the number of arguments must be the same as the number of traits selected and must be in order of increasing trait number. With the covariance statements, the number of arguments must be the same as the number of pairs of traits selected. See multivar segregation model parameters, for discussion of the order of arguments.

multivar can also fit a purely environmental model with no genetic component. The `fit environmental model' statement tells multivar to fit a purely environmental model, with no genetic variance. This null hypothesis model is produced in addition to the genetic/environmental model.

The final two statements specify the number of EM iterations and how often the EM estimates are printed out.

Note that one has the opportunity to provide predetermined eigenvalues of the G-matrix of observed individuals. The `set eigenvalues' statement is used to specify the eigenvalues, with the number of values equal to the number of observed individuals. If desired, the eigenvalues can be provided through an input file accessed with a `input eigenvalue file' statement in the parameter file, or through the command line. (See multivar computational parameters).