BLUPF90 family of programs is a collection of software in Fortran 90/95 for mixed model computations in animal breeding. The goal of the software is to be as simple as with a matrix package and as efficient as in a programming language. For general description, see a paper from the CCB'99 workshop, or see a paper on BGF90 at 7th WCGALP.
For variance component estimation, the family offers choices for simple and complicated models; see paper Reliable computing in estimation of variance components. From 2009 the programs are successively modified for genomic selection using a single-step approach (or ssGBLUP) by Ignacio Aguilar and Shogo Tsuruta.
The family contains:
Modules
- SPARSEM
- sparse matrix manipulation
- SPARSEOP - sparse matrix operations including factorization
and inversion
- DENSEOP
- dense matrix operations (T. Strabel)
- PROB - probability routines for use in threshold models and
Gibbs sampling
- GIBBS - operations useful for data manipulation in Gibbs
sampling
Application programs
They
support mixed models with multiple-correlated effects, multiple animal models
and dominance.
o
BLUPF90
- BLUP in memory
o REMLF90
- accelerated EM REML
o
QXPAK -
joint analysis of QTL and polygenic effects (M. Perez-Enciso)
o
AIREMLF90
- Average Information REML (S. Tsuruta)
o
AIREMLRES
- as above with support for heterogeneous residual variances (T. Druet)
o
CBLUP90
- Solutions for bivariate linear-threshold models
o
CBLUP90THR
-as above but with thresholds computed and many linear traits (B. Auvray)
o
CBLUP90REML
- as above but with quasi REML (B. Auvray)
o
GIBBSF90
- simple block implementation of Gibbs sampling
o GIBBS1F90 -
as above but faster because mixed model equations created only once
o
GIBBS2F90
- as above but with joint sampling of correlated effects
o GIBBS3F90 -
as above with support for heterogeneous residual variances
o
POSTGIBBSF90
- graphical tool for post-Gibbs analysis (S. Tsuruta)
o
THRGIBBSF90
- Gibbs -sampling for any combination of categorical and linear traits (D. Lee)
o
THRGIBBS1F90
- As above but simplified (S. Tsuruta)
o
RENUMF90
- a renumbering program that also can check pedigrees and assign unknown parent
groups; supports large data sets
Available by request
o MRF90 -
Method R program suitable for very large data sets; contact Tom Druet.
o COXF90 -
Bayesian Cox model - contact Juan Pablo Sanchez (JuanPablo.Sanchez@irta.cat)
o
BLUPF90HYP
- BLUPF90 with hypothesis testing (F and Chi2 tests) - contact Juan Pablo
as above
Available only under research agreement
o BLUP90IOD - BLUP by iteration on data with support for very
large problems (S. Tsuruta)
o BLUP90IODTHR
- as above for threshold-linear models
o
ACCF90
- approximation of multiple-trait accuracies
o
BLUP90ADJ
- Data preadjustment tool
Included in application programs
o PREGSF90 –
genomic preprocessor that combines genomic and pedigree relationships (I.
Aguilar)
Documentation
Some documentation is available as README files in source directories. See comments on properties of some of the programs. See Shogo Tsuruta pages for FAQ and individual comments on each program. Those interested in programming can access notes from a computing class in Fortran 95.
Condition of use
The programs are free for research but their use should be acknowledged in publications.
Distribution
Sources
Older versions are available as Fortran 95 package for download and compilation on Unix systems. Various problems were reported using free Fortran compilers. At this time, the best compiler appears to be from Intel; it is free for noncommercial use in Linux.
Linux and Mac binaries and Windows executables
Binaries for Linux, Windows and Mac are here. File blupf90_win.zip contains Windows executables + GNUPLOT; an included script file sets access to all binaries and the graphing package.
Support
Limited assistance is provided by Shogo Tsuruta, Ignacio Aguilar, Ignacy Misztal. Genomic support is mainly by Yahoo group blupf90 – sign up at http://tech.groups.yahoo.com/group/blupf90/. For more extensive assistance consider a visit to UGA.