NEWS.md

In nadiv: (Non)Additive Genetic Relatedness Matrices

2.17.2

DEPRECATED

• pin() does not work with asreml version 4 (should still work with asreml version 3 model objects)
  • nadiv will not support this in the future as asreml v4 has vpredict()

NEW

• proLik4(), essentially the same as proLik(), but works on asreml v4

  • proLik() is kept to retain compatibility with asreml v3 model objects

• makeM() creates mutational effects relatedness matrix M directly

  • can be used with brms/JAGS etc. that require relatedness matrices (not their inverse) in mixed models

Small changes

• fixed deprecated use of Matrix non-virtual subclasses

  • addresses issues with Matrix 1.4-2 and specifically >=1.5-0

• new c++ routine to calculate coefficients of inbreeding and D of Cholesky decomposed A matrix

  • follows Meuwissen and Luo's (1992) algorithm
  • standardizes this code and consolidates to 1 location, instead of being spread out as copies in several other places

2.17.1

Small changes

• fix bug in prepPed() as often encountered/reported from use in optiSel
  • see issue

2.17.0 Released to CRAN 14 January 2021

NEW

• makeMinv() creates the inverse of the (additive) mutational effects relatedness matrix.
• makeT() creates the lower triangle of the cholesky factor of the additive genetic numerator relatedness matrix.

Small changes

• update way create matrices to make 'dsCMatrix' from sparseMatrix() instead of Matrix()
  • fixes error caused by change in Matrix package 1.3-0
• update ggcontrib() to internally use makeT() to directly create a subset of the T matrix (which the subset is Q for the special setup for genetic groups).
  • replaces method that created entire T-inverse then invert it to obtain entire T before subsetting all that is needed for Q
• drop 4-number version information down to just 3 number versions

2.16.2.0 Released to CRAN 20 October 2019

NEW

• geneDrop() conducts a gene dropping simulation down a user-supplied pedigree.

2.16.1.0

NEW

• simGG() now simulates phenotypes with heterogeneous additive genetic variances among the genetic groups (i.e., immigrant and resident groups)
  • The change to this simulation function now ensures phenotypes and underlying breeding values are consistent with the mixed model genetic group analysis approach described by Muff et al. 2019. Gen. Sel. Evol.
  • Breeding values have now been "split" to track resident-specific and immigrant-specific breeding values
  • Users should interact with the function the same way as always, as no changes to the function arguments have been made.
• makeGGAinv() added as a new function to construct genetic group-specific inverse relatedness matrices (Ainv).
  • implements the approach in Muff et al. 2019. Gen. Sel. Evol.
  • An example is given in the help documentation (in R, run ?makeGGAinv), but below is a basic example:

ggPed <- Q1988[-c(3:7), c("id", "damGG", "sireGG")]
AinvOut <- makeGGAinv(ggPed, ggroups = 2)$Ainv  #<-- list with 2 Ainv matrices

• added makeTinv() and makeDiiF() functions
  • These create items used in the Cholesky factorization of a relatedness matrix (or its inverse) and/or the individual coefficients of inbreeding f
  • In particular, these are used to construct genetic group specific inverse relatedness matrices, and are used "under the hood" in makeGGAinv().
  • makeDiiF() creates the D matrix of the Cholesky factorization of the relatedness matrix below (i.e., A and the coefficients of inbreeding (diagonals-1 of A)
  • makeTinv() creates Tinv of the Cholesky factorization of the inverse relatedness matrix below (i.e., Ainv)
    • A= T' D T
    • Ainv=Tinv' Dinv Tinv
  • Note, because D and Dinv are diagonal matrices, Dinv= the element-wise operation of 1 / d_ii
    • Consequently, obtaining Dinv from D is trivial
    • Simply do Dinv <- D followed by Dinv@x <- 1 / D@x
  • Users only need to supply a pedigree and the functions do the rest. For example:

makeTinv(Mrode2)
makeDiiF(Mrode2)

2.16.0.1

NEW

Small changes

• update to simPedDFC() to allow more flexibility in designing pedigrees

2.16.0 Released to CRAN 5 May 2018

NEW

• roxygen2 documentation
• Return diagonal of Mendelian sampling variance matrix in makeAinv() and makeS()
  • These (or their inverses?) can be used in JAGS or BUGS when running a quantitative genetic mixed model

Small changes

• default action is to calculate log-determinant of matrices
  • switched from not calculating this by default

2.15.0

NEW

• Functions to construct sex-chromosomal dominance relatedness matrices
  • makeSd() and makeSdsim()
    • These are similar to what makeD() and makeDsim() accomplish for autosomes
    • The output contains the Sd and Sdsim dominance relatedness matrices
    • The inverses of these can be obtained from Sdinv and Sdsiminv and used in a mixed model

Small changes

• proLik() improved/bug fixed to find confidence limits
  • previously would declare confidence limits found when they hadn't been
    • this was due to optimize() quitting too early with default tol argument
  • returns NA if confidence limits are not, in fact, found (e.g., for boundary parameters, variances that are not significantly greater than zero)
  • plot.proLik() now includes vertical lines to better visualize CIs
• use lower_bound algorithm for matrix lookup within c++ code
  • based on c++ std::lower_bound
  • affect makeAinv() and makeD()
  • greater speedup as A^-1 and D become more dense
• create default and class 'numPed' methods for genAssign() and prunePed()
  • can greatly trim down genAssign.numPed() code (and to some extent prunePed.numped())
  • this speeds up/uses less memory
  • since genAssign() and prunePed() are frequently called in many nadiv functions which operate on class 'numPed', this will have modest, but significant performance increases
  • thanks to profvis for bringing my attention to this!

2.14.3 Released 20 April 2016

NEW

• Fuzzy classification of genetic groups to construct A^-1.
  • Allows individuals' phantom parents to be assigned to genetic groups with a probability. Meaning, they can be assigned to more than one genetic group.
  • To implement, the pedigree must have phantom parent identities as unique rows and a matrix of probabilities of group membership for every phantom parent in every genetic group has to be supplied to the fuzz argument.
  • Examples can be seen in the makeAinv.Rd help file or by running the following commands in R:

?makeAinv              # launches the help documentation
example(makeAinv)      # runs the examples in the help documentation

* Notably, fuzzy classification can be set to 'null', where each phantom parent is assigned to one genetic group with probability=1. This produces the same **Astar** matrix as regular genetic group methods (without fuzzy classification). See this demonstrated in the examples of the help documentation.

• Add the makeAstarMult() function to create the inverse numerator relationship matrix with genetic groups (and possibly also fuzzy classification of genetic groups) through matrix multiplication instead of using direct algorithms to set this up.
  • Uses ggcontrib() and makeAinv() to create Q and A^-1 directly, then multiplies these in such a way as to obtain Astar.
  • Examples using the two different types of pedigree formats and either with or without fuzzy classification can be seen in the makeAstarMult.Rd help file or run them in R with the command:

?makeAstarMult      # launches the help documentation
example(makeAstarMult)  # runs the examples in the help documentation

• Add the F2009 dataset
  • This dataset can be used as an example for fuzzy classification of genetic groups when constructing a numerator relationship matrix with groups (i.e., with makeAinv())
  • See a description in F2009.Rd or in R type:

?F2009

• Add the simGG() function to simulate pedigree and phenotype when immigration occurs in a focal population

  • Allows fairly fine control over a simulation. For example, the function is flexible in the: population size, number of immigrants per generation, number of generations, and both spatial and temporal trends in both focal and immigrant populations.
  • This is the function used to simulate the new ggTutorial dataset (below)

• Added the ggTutorial dataset

  • This is a simulated dataset to be used in analyses with genetic group animal model methods.
  • See a description in ggTutorial.Rd or in R type:

?ggTutorial

• LRTest() is now an exported function to do log-likelihood ratio tests

Small changes

• new S3 generic and methods for makeAinv().

  • method dispatch is based on class of the fuzz argument
  • if fuzz == NULL then dispatch the method makeAinv.default()
  • if fuzz == "matrix" | fuzz == "Matrix" then dispatch makeAinv.fuzzy()

• fix issue with proLik() and the confidence interval estimation

  • use LRTest() as basis of constrainFun() within proLik() so consistently define log-likelihood ratio test statistics
  • close issue #4 with commit 978ad610198398848d97e90c4eb57f4834a4c278

2.14.2 Released 5 Feb 2016

New

• ggcontrib() can now incorporate fuzzy classification of genetic groups
  • To facilitate this, the examples for ggcontrib() have been changed. For more information and examples, read the help documentation ggcontrib.Rd or in R type:

?ggcontrib      # launches the help documentation
example(ggcontrib)  # runs the examples in the help documentation

Small changes

• fixed ordering of f coefficients returned by makeAinv()

2.14.1 Released 22 July 2015

2.14.0 Released 3 July 2015

New

• makeAinv() now can construct the augmented A-inverse matrix for genetic groups
  • This change has introduced new arguments to makeAinv(), however, the defaults are set to produce the normal A-inverse. For more information and examples, read the help documentation makeAinv.Rd or in R type:

?makeAinv              # launches the help documentation
example("makeAinv")    # runs the examples in the help documentation

• Improved algorithm underlying makeAinv() - significant speed-up!
• Created new class numPed for pedigrees constructed by numPed().
  • Methods for checking (is.numPed()) and re-ordering rows (ronPed()) currently available
  • To re-order the rows of an integer pedigree of class numPed, use ronPed() instead of typical subsetting operators (e.g., '[') to retain the class attribute numPed. For example:

nPed <- numPed(Mrode2)
is.numPed(nPed) # TRUE
# re-order using typical R functions
nPed_sub <- nPed[order(nPed[, 2], nPed[, 3]), ]
  is.numPed(nPed_sub) # FALSE; see help via ?'[' about dropping attributes
  class(nPed_sub) # matrix
nPed_subnadiv <- ronPed(nPed, order(nPed[, 2], nPed[, 3]))
  is.numPed(nPed_subnadiv) # TRUE
  class(nPed_subnadiv) # numPed

• Re-made (i.e., re-simulated) the warcolak dataset.
  • Codes specifying the sex are now "M" and "F" instead of 0 & 1.
  • New columns added to the dataset that contain all random effects underlying the phenotype.
  • Entire code used to simulate the dataset is now an example in warcolak.Rd.
• Added two new datasets/example pedigrees: (1) Q1988 from Quaas 1988 and (2) Mrode3 from Mrode (2005) chapter 3. See their descriptions in Q1988.Rd and Mrode3.Rd or in R type:

?Q1988
?Mrode3

Small changes

• Removed the default from the heterogametic argument in makeS().
  • Now this has to be specified by the user!
  • For example:

makeS(FG90)

won't work! Must now type:

makeS(FG90, heterogametic = "0")

as a minimum

2.13.3 Released 4 June 2015

New

• Added TDtT(), a function to take the TDT' Cholesky decomposition of a matrix (not currently exported).
• Added founderLine() which traces all individuals back to either the paternal or maternal founder
• grfx() now has a new argument to allow user to supply the standard normal deviates instead of generating them within the function.
  • extended the warn argument to apply to the warning when incidence = NULL
  • updated grfx.Rd with an example illustrating the stdnorms argument
  • added ... argument to drfx() so that arguments for the internal use of grfx() can be supplied to drfx().
• argument now allows specified prefix for all identities in a pedigree generated from simPedHS() or simPedDFC().
• argument added that specifies output format of ggcontrib(), default is "matrix"

Small changes

• removed 'asreml' from suggests in the package DESCRIPTION file.
• changed pcc() return FALSE if the object (asreml) shows the log-likelihood did not converge
• added silent = FALSE argument to pcc() so that the default can be changed to not show messages
  • helpful in simulations where a lot of output would be printed on screen
• changed the signs associated with likelihood ratio test statistics, etc.
  • changed signs in proLik() so that profile likelihoods should be "valleys" (instead of "hills", as they were in versions previous to 2.13.3)

2.13.2 Released 20 June 2014

Small changes

• added the calculation of the log determinant of the A matrix to makeAinv()
  • log(det(A^-1)) = log(1) - log(det(A))
  • uses property of determinants that det(A^-1) = 1 / det(A) = det(A)^-1.
• changed methods underlying makeA() to use inverse of cholesky factorization of the A-inverse matrix
  • base::chol2inv() to obtain A
  • informally seems faster unless A is dense.

2.13 Released 16 June 2014

New

• added the prepPed() to prepare pedigrees for use in other functions
• exported makeAinv()
• added ggcontrib() so that genetic group contributions can be calculated
  • still need to implement functionality for "fuzzy classification"
• support for selfing
• added the pin() and pcc() functions for the delta method and parameter value convergence checking, respectively, for asreml type REML models. Also added the pin.Rd and pcc.Rd help/documentation files.
  • NOTE, pin() is not exported in this version (need nadiv:::pin() to use it)
• added makeDufam(), but did not export it.
  • experimental version of makeD that first sorts individuals according to generation and then dam, and then sire.
  • sticks individuals with the same parents next to each other in the pedigree
  • haven't implemented a parallel version of this c++ code yet (or checked function for timing/memory benefits or accuracy).

Small changes

• changed makeDomEpi() argument "Dinverse" to "invertD" to be similar to makeD()
• added pcc() checks to constrainFun() so that only likelihood ratio test statistics of the constrained model returned if both the loglikelihood & parameter estimates have converged

2.9

• enabled parallel processing (forking, so no Windows compatibility) in:
  • findDFC(), makeD(), makeDsim()

2.8

• changed name of FindDFC() to findDFC()

nadiv documentation built on Dec. 8, 2022, 1:11 a.m.