USER-FRIENDLY SOFTWARE

Transcription

1 USER-FRENDLY SOFTWARE for large genetic evaluation systems Manual V

2 MiXBLUP User s Guide MiXBLUP, the Mixed-model Best Linear Unbiased Prediction software for PCs for large genetic evaluation systems This manual is for MiXBLUP version 2.0, builds , released in January 2016 MiXBLUP is developed jointly by LUKE National Resources nstitute Finland and ABGC of Wageningen UR Livestock Research. Authors: J. ten Napel M.P.L. Calus M. Lidauer. Stranden E. Mäntysaari H.A. Mulder R.F. Veerkamp ABGC of Wageningen UR Livestock Research P.O. Box AH Wageningen The Netherlands LUKE National Resources nstitute Finland F Jokioinen Finland More information on 2

3 Contents 1 NTRODUCTON Overview Manual System requirements 6 2 HOW TO START nstalling MiXBLUP software MiXBLUP Licenses License key Example files 8 3 NPUT FLES Data file Pedigree file Covariance components files Covariate table file for polynomial regression models nbreeding coefficient file External inverse relationship matrix file nput files for calculating the inverse genomic relationship matrix within the MiXBLUP parser 15 4 NSTRUCTON FLE Title Data file Pedigree information Covariance component files Covariate table file for polynomial regression models The statistical models Analysis, Solving and Output Options Using a different work directory Using a genomic relationship matrix 35 5 RUNNNG MiXBLUP Starting a MiXBLUP evaluation Choosing a breeding value evaluation or a reliability calculation A breeding value analysis with previous solutions as starting values Monitoring and checking the process nterrupting a process of the kernel 44 6 OUTPUT FLES Solution files Log files Temporary files Reserved filenames 48 3

4 7 TUNNG MiXBLUP Flow of programs Variance covariance matrices not positive definite Convergence problems Optimisation of memory and time 50 8 REFERENCES 51 9 ACKNOWLEDGEMENTS 52 APPENDX. EXAMPLES OF ANALYSES N MiXBLUP 53 Example A 54 Example B 55 Example C 56 Example D 57 Example E 58 Example F 59 Example G 60 Example H 61 Example 62 Example J 64 Example K 66 Example L 67 Example M 68 Example N 69 4

5 1 NTRODUCTON MiXBLUP 2.0 has been developed for routine breeding value estimation in commercial genetic programmes and supports modern applications, such as random regression models, group selection, the use of genetic markers or haplotypes and the use of a genomic relationship matrix. 1.1 Overview Development of the software MiXBLUP was initiated to utilize efficient computing strategies for solving mixed model equations. With MiXBLUP it is possible to use sophisticated models in estimation of breeding values in animals, like cattle, pigs, poultry, sheep, horses etc. The MiXBLUP software also includes gene- and markerassisted genetic evaluation using best linear unbiased prediction (BLUP), which is currently the common methodology for genetic evaluation. The software was initially developed for classical genetic evaluation without the use of markers or genes by LUKE National resources nstitute Finland. The adaptation for gene- and marker-assisted genetic evaluation was implemented by Wageningen UR Livestock Research in collaboration with LUKE. The MiXBLUP parser and kernel have been developed for efficient use of disk space and memory. Due to iteration on data and a very fast algorithm in the kernel (preconditioned conjugate gradient), MiXBLUP is able to solve mixed model equations very fast. 1.2 Manual This manual will guide the user through the use of MiXBLUP. The examples provide a way to test MiXBLUP, to get a feel for the software. A general example is used throughout the manual as an illustration for the various input files and models that can be used. Another set of examples are provided as an Appendix to the manual. A schematic overview of the input files, output files and instruction file is in Figure 1. 5

6 Figure 1. Schematic overview of the input and output files of MiXBLUP. 1.3 System requirements MiXBLUP is written in standard Fortran 90 language and is self-contained. The program runs in Windows, Linux and Unix environments and is available in 32-bit and 64-bit versions. n Windows, MiXBLUP runs in the commandline interpreter, cmd.exe (DOS box). With respect to Windows it is tested under Windows XP, Windows Vista and Windows 7. MiXBLUP allocates memory depending on the need. Small applications can be run with a minimum of memory available. Very large applications may need a substantial amount of memory, especially the calculation of reliabilities. For a reliabilities analysis, the user can increase memory allocation with the!maxnonz qualifier in the SOLVNG section (Chapter 4.7). 6

7 2 HOW TO START 2.1 nstalling MiXBLUP software Download the appropriate zip-file from and unzip the folder with the executables: reliabilities.exe, dataprocessor.exe, solver.exe and MiXBLUP.exe in the work directory of your choice for Windows. For Linux the software needs to be installed in the bin directory. Alternatively, the four executables may be installed in a central folder that can be accessed from other folders. f a central folder is used, the user needs to create a file, named SysDir.inp, which contains the path to the executables. This file should be copied to any folder from which MiXBLUP is run. The path to MiXBLUP.exe should be included in the command file that starts up the analysis. MiXBLUP uses SysDir.inp to locate the other three executables. 2.2 MiXBLUP Licenses To run MiXBLUP software on your computer you need a license. There are different license types for MiXBLUP. A license can be ordered at The trial license can handle complete datasets and will provide a maximum of 1000 solutions. This will give the user an opportunity to test the software and decide if it suits their needs. The small commercial license can be used for up to 1 million animal equations. This means that a single trait evaluation could be performed with up to 1 million animals in the pedigree. With multi-trait evaluations (n traits) the number of animals in the pedigree can be 1 million/n. The small commercial license does not allow the use of a genomic relationship matrix. The full commercial license, has no limit on the number of animal equations and provides access to all functionality that is commercially available in MiXBLUP. The license key of the commercial licenses is computer-specific. Therefore, if executables and the license key LCENSE.DAT are moved to another computer, MiXBLUP will give an error message. Running MiXBLUP with the run-time option Dl (minus, uppercase D, lowercase L) writes the host name, license type and expiry date in the license file to the screen output. So if you want to transfer the MiXBLUP software with an existing license to a new computer, you must request a new license from info@mixblup.eu with the LCREQST.DAT attached (how to generate a LCREQST.DAT file see below). You will receive a new license for the remainder of the license period. Table 1. The characteristics of the different license types of MiXBLUP. License types License Time limit Limitations Trial License Not computer specific 1 month 1000 solutions Small commercial License Computer specific 1 year 1 million animal equations, pedigree relationship matrix only Full commercial License Computer specific 1 year Unlimited 7

8 2.3 License key The license key provides the information about the MiXBLUP version, the license type and the expiry date of the license. A trial license can be used for one month and a trial license key is not computer-specific. The small and full commercial license can be used for one year. The license key for these licenses is computer-specific. Trial License Order a trial license at After receiving your order, we send the necessary license key to the address stated in the order. Commercial licenses Order a commercial license at While entering the order you are asked to upload one or more LCREQST.DAT files. For each computer you need to upload a separate LCREQST.DAT file. This file is required to generate a license key for your computer. Generating a LCREQST.DAT file and installing the license LCENSE.DAT > Run MiXBLUP.exe once without the need for an instruction file. MiXBLUP creates the file LCREQST.DAT in the working directory. > After payment of the license one or more LCENSE.DAT files will be sent back and should be saved in the bin folder of the corresponding computer(s). > Store the license key LCENSE.DAT in the C:\MXBLUP\bin-folder for Windows and in the /usr/bin-folder for Linux. Alternative license directory f the license key cannot be stored in the default directory, the user may create a file, named LicDir.inp, which contains the path to the license file. f this file exists, MiXBLUP will look for the license file in the specified folder. 2.4 Example files n the MiXBLUP package, five example files are enclosed, which relate to the general example used throughout this manual. These files can be used to test if the software is running correctly and to get a feel for the program and the set-up of the files. The files should be saved in the work folder. Below a short description of the files is given. mixblup.inp instruction file for breeding value estimation datafile.txt data file with phenotypes and effects pedfile.txt pedigree file para.dat (co)variance components for the used random effects ibd.dat inverse identity-by-descent-matrix 8

9 3 NPUT FLES The following files need to be present when starting MiXBLUP > a data file with the data to be analysed > a file with relationship information; this may be a pedigree file, a pedigree file with an inbreeding file or a file with an external inverse relationship matrix based on pedigree information, genomic information or both. > a trait (co)variance components file with trait variance and covariance components for the random effects > an instruction file with file names, record layouts, field types, statistical models and solver options > one or more optional files depending on the model used The input files are illustrated with the general example of two body weight traits measured at two different ages. Various models are used for breeding value estimation to show the possibilities of MiXBLUP. 3.1 Data file The data fields (animals/ effects/ traits) each have their own column in the data file. The data file must be provided in space-separated format, which means that any two columns are separated by at least one space. Data fields can be integer values, alphanumeric labels or real values (i.e. to be read with a decimal point). Details of the data file: > The maximum column width in the data file is 25 characters. > The maximum record length of the data file is 5,000 characters. > When data is alphanumeric, any of the symbols on the keyboard can be used, including a slash ( / ). > An alphanumeric string must not contain spaces or it will be interpreted as two strings. > A class effect, regardless of whether it is declared as integer or alphanumerical, must not be zero or negative if it is a number. Data records with a class effect in the model that is zero are omitted from the analysis by the kernel. Data records with a negative class effect give an error. > The default missing-value indicator for traits and covariates is zero. Data records with a covariate in the model that is equal to the missing-value indicator are omitted from the analysis by the kernel. f zero is a valid level for one of the covariates in the model, another missing-value indicator should be used. Example. Columns in data file: animal D, mean, herd, sex, dam D, haplotype 1, haplotype 2, common environment, pen mate 1, pen mate 2, age 1, age 2, genotype, body weight at age 1, bodyweight at age 2. A A A12 A A A A11 A A A A11 A A A A15 A A A A14 A A A A14 A A A A18 A A A A17 A A A A17 A

10 3.2 Pedigree file The pedigree file consists of the animal identification number (D) and the Ds of their sire and dam in the first three columns. The columns must be separated by at least one space. f a BLOCK variable is added to the pedigree, it must be the fourth column. The pedigree file may contain other information in any number of additional columns, as long as the number of columns is the same for all records. Any animal occurring in the data file, regardless whether with a record or as a maternal, paternal or pen mate effect (in case of a social interaction model), must be present in the pedigree file. Any animal that does not appear in the data file, but exists as an ancestor in the pedigree file must also have its own record in the pedigree file. Animal Ds in the pedigree file must be of same type as the animal D in the data file (either numeric or text). The phantom parent groups must be coded as negative integers. f no phantom parent groups are used, then missing sires and dams must be coded with a zero (0). When marker-assisted BLUP is performed and one is interested in both the QTL-haplotype effects per animal and a total EBV, then both the data and pedigree file should contain the haplotypes for each animal. With multiple QTL in the model the order of the haplotypes in the model must be the same as the order of the haplotypes in the data and pedigree file. Example. Columns in pedigree file: animal D, sire D, dam D. A1 0 0 A2 0 0 A3 0 0 A4 0 0 A5 0 0 A6 0 0 A7 0 0 A8 0 0 A9 0 0 A A11 A1 A6 A12 A2 A6 A13 A3 A7 A14 A4 A7 A15 A5 A8 A16 A1 A8 A17 A2 A9 A18 A3 A9 A19 A4 A Covariance components files Trait covariance components file or parameter file The trait (co)variance components file contains the between-trait variance-covariance matrices of random effects in the model. There are two options for the format of the file: (1) in lower-triangular-matrix form and (2) in sparse-matrix form. 10

11 Lower triangular matrix The lower-triangular-matrix form is the default option and strongly recommended. n this form, the trait covariance components file can be specified as a lower-triangular matrix using trait names to identify the components. This is the most user-friendly way. The name of the random effect is given at the top of the matrix and the names of the traits are given at the start of each line of the matrix. > The lower triangular matrices and the traits within a matrix can be specified in any order. t means that the order given in the section of the instruction file is not leading. > Models should include at least a genetic effect and a residual effect and so does the parameter file. > The number of traits in the matrices can be larger than the number of traits specified in the model section. Only the lines for which the name has been specified in the model section will be used. > The order of the column names must be the same as the order of row names, so variance components are on the diagonal. > Restriction: in case of a marker-assisted BLUP model with the use of haplotype variance-covariance matrices, each matrix needs to be named and numbered, e.g. GV1, GV2, etc. The name GV refers to the use of the General nverse Variance (GV) function in the model. The order of matrices must be the same as the order of haplotypes given in the model lines of the instruction file. See Example in the Appendix. > n the case of multiple genetic effects (e.g. animal, dam, mate), it should be specified immediately after the trait name and within brackets whether it is the genetic variance of animal, dam or mate. > n case of non-genetic random regression, the name of the class effect is specified at the top of the matrix and a line for each combination of trait and the full random regression term in the model of the trait should be specified (see example F in the Appendix). The syntax in previous versions of MiXBLUP with a separate matrix for each random regression term is still supported, but not recommended, as it ignores covariance components between different random regression terms of the same trait. > f the model contains genetic random regression, then all regression terms should be specified (e.g. animal*covar1 and animal*covar2). See example F in the Appendix for a genetic random regression analysis. > f a covariate table file is used for random regression, then the columns should be referred to as cvr00 for the first covariate column, cvr01 for the second column and so on. The name should be lowercase: the use of CVR00 will give an error. > n case of a social interaction model, with multiple mate effects in the model, the first group mate effect in the model should be specified (e.g. mate2*mate2_x). See example H in the Appendix. Example. The lower triangular trait (co)variance components file with two traits (body weight 1 and body weight 2) for non-genetic random regression, animal genetic and residual effects. sex bw1(sex*age1) 100 bw2(sex*age1) bw2(sex*age2) G bw1(animal) 3000 bw2(animal) residual bw bw

12 Sparse matrix format n the sparse matrix form, the order of the matrices must be the same as the order of random effects in the model, with the restriction that the genetic effect should be the last random effect in the model and the elements of its (co)variance matrix should appear in the sparse matrix file just before the elements of the residual (co)variance matrix. The residual (co)variance matrix should be specified at the end of the sparse matrix file. n summary, the order of matrices is: > Non-genetic random effects in the same order as specified in the model > Genetic effects as specified in the model > Residual effect The matrix elements must be specified as the random effect number, row number, column number and the value of the (co)variance. To avoid mistakes, it is recommended to provide the elements of the lower triangle of the matrix, in other words, any column number is smaller than or equal to the row number. Off-diagonals only need to be specified if they are non-zero. When haplotypes are used in the model for marker-assisted BLUP with the use of an inverse BD matrix, both haplotypes are counted as effects, but the same variance components are used for the first and the second haplotype, when haplotypes are combined with the AND function, so the variance components should not be repeated for the second haplotype. Effectively, the effect number corresponding to the second haplotype is skipped from the list of inverse matrix elements. See Example in the Appendix. Example. The trait (co) variance components file in sparse-matrix format with two traits for the animal genetic and residual effects Columns: random effect number, trait row number, trait column number and variance or covariance component #animal #residual Additional inverse variance-covariance matrices of haplotypes Random effects may have a different variance-covariance structure than simply a diagonal matrix or the A-matrix as for additive genetic effects. Examples of other correlation matrices are BD-matrices of haplotypes for markerassisted breeding value estimation or a dominance relationship matrices. These matrices cannot be made by MiXBLUP, but can be read by MiXBLUP as an external inverse variance-covariance matrix. The file with the elements of the inverse variance-covariance matrix should contain all non-zero elements and be constructed as: haplotype D of row, haplotype D of column, inverse matrix element. The order and numbers used as row and column numbers should correspond to the haplotype numbers used in the data file. Haplotype Ds must be integer. The example below gives the inverse BD matrix for the general example with only two haplotypes. 12

13 Example. The inverse variance-covariance relationship matrix (inverse BD matrix) of 2 haplotypes that have a relationship of 0.25 amongst each other). Columns: haplotype D of row, haplotype D of column, inverse matrix element Multiple residual variance-covariance matrices The residual variance may not be the same for all observations. f this is the case, observations can be grouped by their residual variance prior to the analysis. The!RESVARCLASS qualifier identifies the column in the data file that links the observation to the correct residual variance matrix. The file contains a matrix for every class in the column with the!resvarclass qualifier. The name of the matrix is Res followed by the class number between brackets. The example below gives the series of residual matrices for a situation with observations being linked to one of three residual variances classes. Example. The residual covariance file with three residual variance-covariance matrices. Res (1) bw bw Res (2) bw bw Res (3) bw bw Covariate table file for polynomial regression models f the relationship between an independent variable and a dependent trait is modelled as an nth order polynomial, a covariate table file with all levels of the independent variable and (n+1) columns of covariates may be used for easy data preparation and syntax of the instruction file. The covariate table file may be created outside of MiXBLUP, it may exist from a previous analysis or it may be created at run-time. The latter is specified by using the!cvrmake qualifier. Currently only Legendre polynomials can be created this way by specifying LEG behind the!cvrmake qualifier. The required order of the Legendre polynomial can be specified behind the!cvrnum qualifier. See 4.5 for details. f the order is n, the covariate columns in the table are numbered from 0 to n, giving n+1 covariate columns in addition to the original independent variable. The independent variable corresponds with a column in the data file, indicated with the!cvrnd qualifier after the column name in the instruction file. This column must contain a valid entry for every data observation. The 13

14 columns in the covariate table file can be specified in the model with CVR(n) or equivalently with cvr00, cvr01,..., cvrn. See for details. Example. A covariate table file for an independent variable with values in the data between 86 and 115. The order of the Legendre polynomial is 2. The table was created with the line CVRTABLE!CVRMAKE LEG!CVRNUM 2!CVRMN 86!CVRMAX 115 in the instruction file E <lines for 94 to 107 omitted> E nbreeding coefficient file The internally calculated numerator relationship matrix (A -1 ) is by default set up without taking into account inbreeding. nbreeding can be taken into account by providing the kernel with a file with the inbreeding coefficient of each animal in the pedigree file. This file may be provided as an input file or calculated within MiXBLUP. This file may contain any number of columns, but at least the original animal D and the inbreeding coefficient of the animal. Example. nbreeding coefficient file with animal D in the first column and the inbreeding coefficient in the fourth column. A A A3 A1 A2 0.0 A A5 A1 A4 0.0 A6 A3 A < > A19 A16 A A20 A16 A

15 3.6 External inverse relationship matrix file The external inverse relationship matrix that replaces the internally calculated numerator relationship matrix (A -1 ) is provided in a file with three columns. The first column is the original D of the animal of the row of the matrix, the second column is the original D of the animal of the column of the matrix and the third column is the matrix element. The original D may be a number or a character string. The file only needs to contain the lower-triangular part of the matrix, but a file with the upper-triangular part or a full-rank matrix will be read correctly. MiXBLUP, however, does not check that the provided matrix is symmetric. The external relationship matrix file may be located in a different folder as the MiXBLUP instruction file, if the path to the file is added to the filename. Example. Columns in external inverse relationship matrix file: animal D row, animal D column, matrix element. A1 A A2 A1-0.5 A2 A2 2 A3 A1-0.5 A3 A A3 A3 2 < > A19 A nput files for calculating the inverse genomic relationship matrix within the MiXBLUP parser The inverse genomic relationship matrix can be calculated from genomic data within the MiXBLUP parser, using the calc_grm software. This analysis requires at least a file with genomic data. For the calculation of the blended relationship matrix, a genomic data file and a pedigree file are required. Optional files are the breed composition file for multi-breed analyses and the allele frequency file for the use of pre-defined allele frequencies instead allele frequencies calculated from the data Genotype file The genotype file contains the genomic data. t must contain the animal D in the first column and genetic marker data from the second column onwards. The animal D and the genetic marker data must be separated by at least one space. The genetic marker data itself may be presented as alleles in pairs, alleles on two lines per animal or genotypes. The format of the file may be either dense (genomic data as one string per animal) or space-separated (at least one space between any two genetic markers). The genomic data cannot be partly dense and partly space-separated. Example. Genotype file with genomic data presented in six pairs of alleles per animal in space-separated format. A A A A A < > A

16 Example. Genotype file with genomic data presented in two lines of alleles per animal in dense format. A A A A A A < > A A Example. Genotype file with marker genotype data per animal in dense format. t contains the number of copies per locus of the allele with the highest number (11=0, 12=1 and 22=2). A A A A A < > A Pedigree file The pedigree file for the calculation of the blended relationship matrix must contain the original animal Ds of animal, its sire and its dam in the first three columns. t may contain additional columns, which are ignored Breed composition file The breed composition file contains the original animal D in the first column and contains a number of additional columns that is equal to the number of breeds specified. The breed composition may be presented as a number, for example 4 (out of 8 or any other number), as a percentage, for example 50, or as a fraction, for example MiXBLUP converts the breed composition of an animal to the value of one breed over the sum of values across breeds. For example in an analysis with four breeds, animal X having as the breed composition will be converted to X t is therefore essential that the breed information is complete, so add a column for unknown or other, if necessary. All columns must be separated by at least one space. Example. Breed composition file with the percentage of four breeds per animal. A A A A A < > A

17 Example. Breed composition file in parts of one eighth of four breeds per animal. A A A A A < > A Allele frequency file f the user does not want to use allele frequencies calculated from the data, then pre-calculated allele frequencies must be supplied as an additional input file, The file specified should contain per locus the allele frequency of the allele with the highest integer code, if the genetic marker file contains alleles. The file specified should contain per locus the frequency of the allele of which the homozygote genotypes are coded as 1. The structure of the file is <locus number in order of the genetic marker file> <allele frequency>. Example. Pre-calculated allele frequency per locus of allele coded as 1 for 6 loci

18 4 NSTRUCTON FLE n the instruction file, the user specifies the names and locations of the input files, the record layout of the data and pedigree file, the statistical models of the analysis and any additional information to control the analysis. n the following paragraphs, the different sections of the instruction file will be discussed and the contents and options for each of those sections will be explained. Below the example instruction file is given for a bivariate animal model for two body weight traits (bw1 and bw2). Example. nstruction file for a bivariate animal model. TTLE breeding value estimation for body weight 1 and body weight 2 DATAFLE datafile.txt animal A mean herd sex dam A haplo1 haplo2 commonenv mate1 A mate2 A age1 R age2 R genotype R bw1 T bw2 T PEDFLE!groups 0.0 animal A sire A dam A PARFLE para.dat bw1 ~ herd sex!random G(animal) bw2 ~ herd sex!random G(animal) SOLVNG!maxit 1000 Details of the instruction file: > The maximum record length of the instruction file is 5,000 characters. > The instruction file may contain empty lines for the convenience of the user > Comments may be inserted on a new line or after instructions on the same line, provided that any comment starts with a hash (#). > The keyword of any section must be the first word of the line. > Sections may appear in any order. 18

19 > Mandatory sections in MiXBLUP are TTLE, DATAFLE, PEDFLE or ERMFLE, PARFLE, and SOLVNG. > Qualifiers within a section may generally appear in any order and anywhere in the section, provided that they are not linked to a specific field in the DATAFLE section or a specific trait in the section. > Qualifiers start with an exclamation mark (!). > A statement may be continued on the next line by using an ampersand (&) as the last word of a line or the last word before a hash (#). > Section keywords, qualifiers, labels, values and the special characters #, & and ~ must be separated by at least one space. > The syntax is described below for each section separately. The <...> is used to indicate a value or text label provided by the user as input. The [...] is used to indicate an optional qualifier or optional input. Keywords for sections and qualifiers are in capitals. The ampersand (&) is used to continue the syntax on the next line. 4.1 Title The instruction file must start with a specification of the title of the analysis. The TTLE keyword is optional. f omitted, the first line must be a comment line, starting with a hash (#). This comment line is then used as the title of the analysis. Syntax: TTLE <description of analysis> Alternative syntax: # <description of analysis> Example. Section TTLE in the instruction file. TTLE breeding value estimation for body weight 1 and body weight Data file The instruction file contains the name of the data file, the specification of fields in a record and the type of each field. The data file is located by default in the work directory, but it can be in any other folder if this is specified as part of the name of the file (e.g. d:\performancetest\breedp.txt). The order of the fields in the DATAFLE section must be the same as the order of the columns in the data file. Syntax: DATAFLE <filename> [!SKP <n lines>] [!MSSNG <value>] <field 1> <field type: /R/T/A>... [<field i>!block]... [<field j>!resvarclass]... [<field k>!cvrnd]... [field n] [/R/T/A] 19

20 Qualifiers:!missing <value> f the value specified for MSSNG is encountered when reading the data file, it is interpreted as a missing observation for the trait or covariate. A missing covariate invalidates the trait for which the covariate is included in the model.!block This field is used as the block variable. f used, the data file and pedigree file both need to contain this column. t is required for the calculation of reliabilities, but might be beneficial in some computationally heavy genetic evaluations. The field must be integer. The!block qualifier must not be specified in the PEDFLE section, but the fourth column in the pedigree file must have the same field name as the block variable in the data file.!resvarclass This field is used to specify the residual variance class of the data record, in case the residual variance differs for groups of records. The field must be integer. The qualifier!resvarclass must be used if the section RESFLE is specified.!cvrnd The field marked with!cvrnd is the independent variable used in polynomial regression. Any level of the field specified with!cvrnd must exist in the covariate table file. The field must not contain a missing value indicator for a valid trait observation. The qualifier!cvrnd must be used when the section CVRTABLE is specified. The field must be integer.!skip [value] With this qualifier, one (!SKP 1) or more (e.g.!skp 2) header lines in a data file can be ignored when reading the data file Example. Section DATAFLE in the instruction file DATAFLE datafile.txt animal A mean herd sex dam A haplo1 haplo2 commonenv mate1 A mate2 A age1 R age2 R genotype R bw1 T bw2 T 20

21 Details of the DATAFLE section: > The field specification must start on the line following the line containing the DATAFLE keyword > The field type indicates whether a field in the data file should be read as an integer value (), a real value for covariates (R), a real value for a trait (T) or a text string (A). > Maximum length of field names is 8 characters. A field name may be up to 19 characters long, but only the first 8 characters are used to distinguish fields, so a warning is given to remind the user. Field names longer than 19 characters result in an error. > Field names are case-sensitive throughout MiXBLUP. > f!block is specified for multiple data fields, only the first specification is used. t affects the SORTED-line in the file generated by the parser (dataprocessor.inp or dataprocessor_rel.inp). > Alphanumeric strings (fields coded with A) are converted into integer values for the analysis. Solutions are decoded to the original alphanumeric levels of the factor. > Each alphanumerical label in a field in the data file gets a unique numerical value. There is no apparent relation between the alphanumerical label and numerical value, so the numerical value of a string may vary across runs without a restart. The numerical value of a string does not change if!restart is specified in the SOLVNG section. > The D of animal in the data file, and the Ds of animal, its sire and its dam in the pedigree file must all be of the same type, so either alphanumeric (A) or numeric (). > The version of the data file with alphanumerical labels converted to integer values is data.txt. > The use of names reserved as section keywords, qualifiers or functions as field names is not supported. 4.3 Pedigree information Pedigree file The instruction file specifies the pedigree file name. The pedigree file is located by default in the work directory, but it can be in another folder if the folder is specified as part of the name of the file (e.g. d:\pedigrees\pedigreebreedp.txt). The field order is: (1) animal D, (2) sire D and (3) dam D. Without using a blocking variable, the pedigree file does not need to be sorted. Optionally, a block variable can be given as column 4. n that case the pedigree file, as well as the data file, should be sorted on the block variable. A sire model with sires and maternal grandsires in the pedigree file is not supported in MiXBLUP. Syntax: PEDFLE <pedigree file> [!GROUPS <value>] [!SKP <n lines>] [!CALCNBR] <field animal> <field type> <field sire> <field type> <field dam> <field type> [<field blocking variable> <>] Qualifiers:!Groups <value> The qualifier GROUPS means that phantom parent groups are included in the pedigree. Phantom parent groups need to be coded with negative integer values. With <value>, it is possible to specify whether these phantom parent group effects should be modelled as fixed (value=0) or as random (value>0.0), the larger the value the more random it is, or the more estimates are regressed towards the mean. n practice,!groups does not need to be set at a much higher value than about 5. 21

22 !SKP <n lines>!calcnbr The SKP qualifier may be used to skip the first n lines of the pedigree file. The qualifier CALCNBR is optional and is used to indicate that inbreeding coefficients should be calculated and included in the calculation of the inverse pedigree relationship matrix (A-1). f!calcnbr has been specified, the section NBRFLE (see 4.3.2) is ignored. The default setting is that inbreeding coefficients are not taken into account when setting up the inverse pedigree relationship matrix. Example. Section PEDFLE in the instruction file. PEDFLE pedigree_breed_a.txt!groups 0.0!CALCNBR Animal A Sire A Dam A HerdBlock Comments: > The D of animal in the data file, and the Ds of animal, its sire and its dam in the pedigree file must all be of the same type, so either alphanumeric (A) or numeric (). > The maximum record length of the pedigree file is 5,000 characters. > Alphanumerical Ds (A) are translated into numerical values. The version of the pedigree file with animal Ds converted to integer values is pedigree.txt. > When marker-assisted BLUP is performed and one is interested in the QTL-haplotype effects per animal and a total EBV, then the pedigree file needs to contain the haplotypes for each animal. With multiple QTL in the model, the haplotypes should be given in the order as given in the model. > The pedigree file may contain additional fields that do not need to be specified. Additional fields will be ignored. The number of additional fields must be the same for every record in the pedigree file. > f the pedigree contains a blocking variable, it must be the fourth column Using an external file with inbreeding coefficients By default, the calculation of the inverse pedigree relationship matrix does not take into account the inbreeding coefficient of individuals. f a file with inbreeding coefficients for all individuals in the pedigree is available, the section NBRFLE can be used to specify this file, indicate the field numbers of animal D and inbreeding coefficients and include the inbreeding coefficients in the calculation of the inverse pedigree relationship matrix. Syntax: NBRFLE <file with inbreeding coefficients> [!DCOL <column number>] [!NBRCOL <column number>] Qualifiers:!DCOL <value>!nbrcol <value> The optional qualifier!dcol can be used to specify the field number in the inbreeding coefficient file that contains the animal D. The default field number is 1. The optional qualifier!nbrcol can be used to specify the field number in the inbreeding coefficient file that contains the inbreeding coefficient. The default field number is 4. Example. Section NBRFLE in the instruction file. NBRFLE inbrcoef_breed_a.txt!dcol 1!NBRCOL 4 22

23 Comments: > The section NBRFLE is ignored if!calcnbr is specified in the PEDFLE section (see 4.3.1). > The section NBRFLE must be specified in conjunction with the PEDFLE section. t is only used by the MiXBLUP kernel for calculating the inverse pedigree relationship matrix (A -1 ) or the blended inverse genomic and pedigree relationship matrix (H -1, see 4.9.3). > The file with inbreeding coefficients may contain any number of fields per record, as long as the number of fields per record is the same for all records. > t is assumed that the animal D in the inbreeding coefficient file is of the same type as in the pedigree file. 4.4 Covariance component files Trait covariance component file or parameter file The instruction file specifies the name of the trait (co)variance components file. The trait (co)variance components file is located by default in the work directory, but can be in another folder if specified in the name of the file. The trait (co)variance components file can be in two different formats. The default format is in lowertriangular matrix form. The other format is the sparse matrix form. The lower triangular form provides great flexibility. Syntax: PARFLE <filename> [!SPARSE] Qualifier:!Sparse f!sparse is specified, the variance and covariance components are read in sparse matrix form. f omitted, the matrix is read in lower triangular form. Example. Section PARFLE in the instruction file. PARFLE para.dat Additional inverse variance-covariance matrix of haplotypes The names of any files with an additional inverse correlation matrix, for example inverse BD-matrices, can be specified in the section CVMATRX. A file with an inverse correlation matrix is located by default in the work directory, but it can be in another folder if the folder is specified as part of the name of the file. The order of files is important and should correspond with the numbers given as the second argument of the General nverse Variance function (GV function). The inverse correlation matrix should contain all non-zero elements and each record should contain: row_number, column_number, element of inverse correlation matrix. The file should not contain alphanumeric data. The inverse BD-matrix should contain at least one off-diagonal. The section CVMATRX can only be used in connection with the GV function in the statistical model (4.6.5). The dimension of the inverse correlation matrix is equal to the number of levels of the random effect, in other words, the number of alleles for a genetic marker or the number of haplotypes for a combination of markers. 23

24 Syntax: CVMATRX <filename1> [<filename2>] [<filename3>] # corresponds with GV(<...>,1) # corresponds with GV(<...>,2) # corresponds with GV(<...>,3) Example. Section CVMATRX in context in the instruction file. CVMATRX BD.giv # corresponds with GV(<...>,1) in the section bw1 ~ herd sex!random GV(haplo1 AND haplo2,1) G(animal) Multiple residual covariance component matrices f multiple residual covariance component matrices are required, then the file containing the matrices should be specified in the section RESFLE. The section does not have any qualifiers. The file is located by default in the work directory, but can be in another folder if specified in the name of the file. The trait must be specified in lower triangular matrix form. Syntax: RESFLE <filename> Comments: > The file must contain a matrix for every level of the column specified as the residual variance class (!RESVARCLASS). Example. Section RESFLE in the instruction file. RESFLE respara.dat 4.5 Covariate table file for polynomial regression models Covariate table files for polynomial regression are specified with the CVRTABLE section. 24

25 Syntax: CVRTABLE <filename> CVRTABLE!CVRMAKE LEG!CVRNUM <nth order>!cvrmn <minimum value>!cvrmax <maximum value> Qualifier:!CVRMAKE f!cvrmake is specified, MiXBLUP generates a covariate table file using the settings specified with the!cvrnum,!cvrmn and!cvrmax qualifiers. Currently, only a covariate table containing Legendre polynomials can be created, by specifying LEG as the argument of!cvrmake. The name of the new covariate table file is cvrtable.txt.!cvrnum The qualifier!cvrnum must be specified and is used to specify the order of the polynomial in the covariate table. The expected number of columns to read is the order + 2, one for the level of the independent variable and one for the order being 0. t is up to the user to make sure that the order specified in the section is equal to or lower than the order specified with!cvrnum.!cvrmn and!cvrmax The qualifiers!cvrmn and!cvrmax can be used to specify the lowest and highest value of the independent variable that were used to estimate the genetic parameters. Legendre polynomials are dependent on the lowest and highest value of the independent variable and so are the genetic parameters of Legendre polynomials. f!cvrmn or!cvrmax is nevertheless omitted, the lowest or highest value of the independent variable in the data is used, instead. Comments: > The filename cvrtable.txt may also be used for an existing covariate table file > The column in the data file that contains the independent variable is marked with!cvrnd in the DATAFLE section. > Regression on a polynomial of an independent variable is either specified using the CVR(...) function or by specifying the individual terms separately. n the latter case, the columns of the covariate table file are named cvr00 to cvrnn, where nn is the highest order of the polynomial in two-digit format. Example. Section CVRTABLE specifying an existing file. CVRTABLE DaysnMilk.txt!CVRNUM 9!CVRMN 14!CVRMAX 308 Example. Section CVRTABLE specifying the creation of a new covariate table file. CVRTABLE!CVRMAKE LEG!CVRNUM 9!CVRMN 14!CVRMAX

26 4.6 The statistical models The section specifies the start of the statistical models for the traits in the analysis. The statistical models start immediately below the line with the keyword. For each trait, the statistical model is specified on a separate line. MiXBLUP supports up to 63 traits to be analysed simultaneously, if the computer resources permit this. Each line starts with the name of trait. t must at least contain a tilde (~) to separate the trait from the statistical model, one fixed effect, the!random qualifier to separate the fixed from the random effects and a direct genetic effect enclosed in the function G(), for example G(animal). MiXBLUP uses either the pedigree to construct the inverse numerator relationship matrix (also known as the inverse pedigree relationship matrix) or the genotype file to construct the inverse genomic relationship matrix or a blended inverse genomic and pedigree relationship matrix for the genetic random effects. The residual random effect does not need to be specified. The following paragraphs discuss the basic model (4.6.1), the maternal genetic model (4.6.2), the random regression model (4.6.3), the social interaction model (4.6.4) and the marker-assisted BLUP model (4.6.5) The basic statistical model The basic statistical model for a breeding value evaluation contains fixed effects and a direct genetic random effect. t may also include fixed regression, nested fixed regression and uncorrelated, non-genetic random effects. See Example A, Example B and Example C in the Appendix for analyses using a basic statistical model. Syntax of the basic statistical model: <trait1> [!WEGHT <weighting factor of trait 1>] ~ <fixed effect> [BL(<fixed effect>)]& [covariates] [nested covariates]!random G(<direct genetic effect>) [<non-genetic random effects>] [<trait2>...]... [<traitn>...] Qualifiers and functions:!weight A field in the data file may be specified as a weighting factor for a specific trait using the!weght qualifier. A possible application is a trait, which is based on a number of observations that varies between animals. The weighting factor could specify the number of observations included in the trait value.!random The!RANDOM qualifier separates the fixed effects and fixed regression from the random effects in the model. BL(<fixed effect>) The BL(...) function may be used to make the analysis computationally more efficient. t places the fixed effect, typically with a large number of classes, within the classes of the block field, as indicated with!block in the DATAFLE section. G(<genetic random effects>) The G(...) function links a random effect to the inverse relationship matrix, constructed from pedigree and genotype information. 26

27 Fixed effects with classes At least one fixed effect with classes must be specified in the model. f there is no fixed effect to include, the user should add a column of ones to the data file and specify this column as the fixed effect. This is the overall mean. MiXBLUP does not support interaction between fixed effects with classes. To fit an interaction, the user should combine the two fields in the data file into one field and fit the combined field in the model Fixed regression To specify a fixed regression, just add the covariate to the model. Fields specified in the DATAFLE section with type R are considered to be covariates. Fields specified with or A before the!random qualifier are considered to be fixed effects with classes. For fitting a polynomial regression, see Chapter Nested fixed regression A fixed regression may be nested within the classes of a fixed effect. This should be specified as <fixed effect>*<covariate> or <covariate>*<fixed effect> as one word without spaces. The regression coefficient is then estimated within each class of the fixed effect Direct genetic random effect One random effect must be linked to the inverse numerator relationship matrix (A -1 ) and this is specified with the G(<random effect>) function. This is typically the animal D that links the pedigree file with the data file Uncorrelated, non-genetic random effects The basic model may also contain additional random effects, for which all classes are uncorrelated. The inverse correlation matrix of classes for any trait is therefore an identity matrix. Example. Section with a basic statistical model. herd is a fixed effect with classes; age2 is fixed regression; sex*age1 is fixed regression nested within sex ; G(animal) is the direct genetic random effect; commonenv is the uncorrelated, non-genetic random effect. bw1 ~ herd sex sex*age1!random G(animal) commonenv bw2 ~ herd sex age2!random G(animal) commonenv The maternal genetic model Some traits are affected by the genotype of the animal itself and the genotype of its dam at the same time. An example is weaning weight in beef cattle. For such traits, a maternal genetic model should be used. The inverse numerator relationship matrix is applied both to the direct genetic effect (animal) and the maternal genetic effect (dam). The maternal genetic effect must be a separate field in the data file and each animal used as a maternal genetic effect must exist as an animal in the pedigree file. For biological dams, this is self-evident, but for foster dams, this requires special attention. See Example D in the Appendix for the instruction file and the trait (co)variance component file for an analysis using a maternal genetic model. 27