WOMBAT – A program for Mixed Model Analyses by Restricted Maximum Likelihood

Updates and changes to WOMBAT

This page tries to keep track of what has been changed in WOMBAT and when — including `bug' fixes.

2014

10/09/2014

Some minor bug fixes. Added option to store *.gin matrix in core.

2013

19/11/2013

Fixed bug affecting runs using the –snap option when multiple SNPs were fitted simultaneously.

4/9/2013

Added: trial versions compiled for Windows (both 32 and 64 bit) using the gfortran cross-compiler on a Linux host. These have all the features of the Linux versions, but had very little testing - beware!

22/7/2013

Added: Example 17 to illustrate approximation of sampling variances for complicated functions of covariances via sampling from a multivariate normal distribution.

19/3/2013

Linux versions updated: Changed sparse matrix computations to take advantage of “supernodes” - this speeds up computations but can require quite a bit more memory. Run time option −−old is provided to switch back to the previous scheme.

The speed up achieved varies greatly with the “quality” of the BLAS and LAPACK routines used - best for the I7 (MKL library) and LF64 (ATLAS library) versions, not quite so good for the PATH64/32 or LF95 versions which are compiled on older machines with only the reference libraries.

1/3/2013

Added run option −−dense; aimed at analyses with a dense, user-defined covariance matrix among random effects (e.g. genomic relationship matrix); not everything works yet; dense calculations rely on LAPACK and BLAS libraries. also new: 64bit version compiled for I7 processor using Intel ifort compiler and MKL library.

2012

15/08/2012

Downloads updated. New: facility to fit “social” genetic effects & Example 16 to illustrate it.

21/5/2012

Berenice H. reported that WOMBAT scrambled the starting values supplied for certain types of random regression models. Fixed.

19/4/2012

Update downloads. New option −−pool: allows pooling of estimates from part analyses using penalized maximum likelihood; illustrated in Example 15.

2011

25/9/2011

Updated Linux versions and examples. New: Example13 & Example14, option TSELECT for RPTCOV, −−nonly for analysis of first n records only, −−nolsq to skip LSQ analysis for fixed effects and −−snap for GWAS analysis.

24/2/2011

Added section on fitting imprinting effects to Specialized analyses; added example4/D to illustrate.

22/2/2011

Revamped options for penalized estimation; updated manual, Linux versions & example5/D.

2010

14/4/2010

Updated linux versions: option MRR enabled (full rank analyses only) – beware, has only been tested for a simple bivariate case. New special option RPTCOV & new option ONE for random regression basis functions.

2009

4/12/2009

New options: −−redped switches on pedigree reduction for sire model; IDE as code for basis functions for random regression models; data subset selection for specified range of values of a covariable.

4/8/2009

John K. noted that fitting correlated random effects did not work for effects with the GIN option; fixed.

5/4/2009

Changed rules for sire model analyses: require “subject” identifier in data & model part of parameter file.

31/03/2009

Romain D. reported a problem for an analysis with many dependencies among fixed effects – increased limit and improved fault trapping for exceeding the maximum number.

27/03/2009

Added checks for sorting of data file; new SPECIAL option CLONES. Analyses with very low (<10% but >0%) proportions of repeated records for a trait are no longer accepted.

21/01/2009

Changes made last year caused WOMBAT to no longer recognise the DIAG option for random effect covariance matrices; fixed. Option ––choozhz was only recognised for animal model analyses; now enabled for sire models as well.

2008

17/10/2008

Christie I. reported that the COVZER option did not work properly for multivariate analyses with numbers of covariates differing between traits. Klint G. found that the Windows versions had problems if data and pedigree file were the same. Both fixed.

Added facility to read in “pre-defined” inbreeding coefficients from the pedigree file.

26/8/2008

Fixed bug which affected convergence of reduced rank analyses under a sire model with known maternal grand sires.

13/8/2008

Fixed bug which caused WOMBAT to ignore some of the effects chosen to be set to zero through the ZEROUT directive, and improved `automatic' identification of dependencies among fixed effects somewhat.

30/7/2008

WOMBAT now stops if it finds repeated records for a trait for less than 20% of individuals, under the assumption that this is due to careless data preparation. For the rare cases were this is a planned scenario, this can be overridden by a new 'special' option (see section 4.8 of the manual). In addition, WOMBAT now checks for duplicate identities in the pedigree file and stops if these give different parental identities.

11/7/2008

When writing out solutions for random effects for multivariate analyses using the “PC” option, WOMBAT now produces additional files with corresponding solutions transformed `back' to the original scale.

1/7/2008

WOMBAT now checks for values of zero for covariables fitted. If the data contains such values, it expects a 'special' option (see section 4.8 of the manual) telling it whether to ignore such records or treat them as valid values for the covariable.

16/5/2008

Added facilities to a) fix selected covariance matrices at their `input' values, b) replace the “1” in design matrices (standard uni- and multivariate analyses only) by a value (weight) obtained from the data file.

15/5/2008

Warren S. pointed out that WOMBAT would not allow correlated random effects in conjunction with a general relationship matrix. Changed - WOMBAT now allows both NRM and GIN options for such effects.

24/4/2008

Added option −−mmeout — this is like −−solvit but writes out the complete mixed model equations to file, including solutions and mapping of effects fitted to equation numbers. This is the option for folks who want to calculate linear contrasts, F-ratios etc. for fixed effects.

8/4/2008

Martin L. reported that WOMBAT would complain about an animal occurring as sire and dam for a sire model when a bull occurred as sire and MGS. Fixed.

1/4/2008

New run option: −−solvit provides a BLUP analysis solving the mixed model equations iteratively.

20/2/2008

Pia B. reported that WOMBAT had trouble coping with blank lines in the pedigree file. Fixed.

8/1/2008

Added option for `stand-alone' matrix inversion.

2007

20/11/2007

Have changed (made faster) computation of average information matrix in reduced rank analyses - Cholesky factor of numerator relationship matrix no longer required or written out (Files cholnrm*.bin).

17/07/2007

David H. reported that the Windows versions failed for multivariate, reduced rank analyses while the Linux versions did not. This turned out to be due to a minor bug in the g95 compiler. Both compiler and Windows versions have been fixed.

29/06/2007

New features have been added :

  1. WOMBAT can now carry out analyses involving subsets of traits in multivariate analyses, without the need to extracts subsets of the data file or renumber traits.
  2. An option is provided to write out the parameter files for bivariate analyses of all pairs of traits, using the parameter file for the complete analysis as input
  3. When encountering such analyses of subsets, abbreviated files of partial results are written out automatically. These can be used as input files when pooling results.
  4. WOMBAT can now combine results from multiple analyses of subsets of traits into positive definite, pooled covariance matrices using a (weighted) iterative summation of expanded part matrices.

Known problems

  • Andrew S. reported that the Windows versons of WOMBAT crash when attempting to write out a “Curve_” file for a nested covariable - this seems to be a problem associated with a '*' in the file name.
  • Occasionally WOMBAT hangs up or crashes during the set-up phase. This occurs mainly for very large data sets or models with a dense coefficient matrix in the mixed model equations, and can be attributed to 'failure' of the routine used to re-order the mixed model equations.
    WOMBAT employs three different public domain routines to carry out the ordering. Unfortunately, none of them is `bomb-proof'. In particular, the routine from the MeTiS package (which is used by default for large analyses) is prone to problems and does not provide the possibility of a graceful exit when this occurs. If this happens, please try one of the other ordering strategies or other MeTiS options.
  • Operating system incompatibility? Steven M. reported a 'segmentation fault' for WOMBAT versions compiled with ifort when running under Fedora. This fault was not replicable for other Linux versions or executables compiled with a different compiler and thus has not be resolved. Beware!
QR Code
QR Code wombat:updates (generated for current page)