Whats New in Release 3

Introduction

These notes relating to ASReml 3 are intended for those familiar with ASReml 2 and describe additions to the syntax. New users should read the User Guide which provides a comprehensive to fitting mixed models in ASReml.

An ASReml Forum is hosted by VSN. Users are urged to join the forum at http://www.vsni.co.uk/forum and participate in the discussion among ASReml users.

Critical Changes to Behaviour

Generally we seek to maintain upward compatability so that ASReml 1 code will continue to run. However, to deliver improved facilities, some changes to behaviour are unavoidable. More.

!ASSIGN statement

The !ASSIGN statement allows a string to be defined early in the job, and substituted in repeated further down.

Expanded testing of fixed effects.

The table of Wald F statistics has been expanded to allow the user to specify the exact conditional Wald F-statistic (!FOWN qualifier) to be tested.

Factor definition: !AS declaration

Diallel experiments are common in some species. The usual way of fitting these in ASReml is to declare the two parents in two factors say Male and Female, and to fit the general combining effect in the model as Male and(Female) so that the design matrix for Female is overlaid on (or added to) the design matrix for Male. This requires that the two factors be coded conformable so that the ith 'male' is the same individual as the ith female. If they are not directly coded that way, a pedigree file can be used, but the new !AS qualifier is more convenient. It is used as in the following example: 
  Male !A
  Female !AS Male
works internally by using a common list of factor level labels the two factors.

Paths and loops in the .as file

The !CYCLE statement has been modified and extended. It can operated in association with PARTs. It no longer needs !JOIN and it expands a integer sequence like 1:100.

Transformations

Dropping records

The !D transformation conditionally discards records depending on the value of its argument and the value in the test field. \rev It drops the record if the test generates TRUE or the the test field is missing. A new form !DV has been implemented which does not drop the the record when the test field is missing unless the test value is *. That is !D >99 is equivalent to !DV * !DV >99.

Use
!DV * to discard records with a missing value in the test field,
!DV v to discard records with a v in the test field,
!DV<v to discard records when the test field has a value <v,
!DV<=v to discard records when the test field has a value <= v,
!DV<>v to discard records when the test field has a value not equal to v, !DV>=v to discard records when the test field has a value >= v, !DV>v to discard records when the test field has a value >v.

Setting the TARGET field

A previous ambiguity with setting the TARGET field has been resolved with the introduction of an explicit !TARGET qualifier.

NA transformation

The !NA v transformation has been extended to allow v to refer to another field instead of just being a value. This facilitates copying a value from another field when the current field has a missing value. e.g. WT1 WT2 WT !=WT2 !NA WT1 defines WT with values of WT2, or WT1 if WT2 is missing.

Pedigree and GIV files

Pedigree file line qualifiers

!XLINK requests an inverse NRM based on the X chromosome be formed along with the usual additive inverse NRM.

!FGEN requests formation of the additive inverse NRM assuming specified levels of inbreeding.

!Goffset o An alternative to group constraints described above is to shrink the group effects by adding the constant o to the diagonal elements of A-1 pertaining to groups. When a constant is added, no adjustment of the degrees of freedom is made for genetic groups. Use !Goffset -1 to suppress adding of constraints where empty groups appear. The empty groups are then not counted in the DF adjustment. There is also extended information on constraints applied to gorup effects.

GIV file qualifiers.

A !GROUP qualifier has been added to associate fixed degrees of freedom to the GIV matrix. The GIV and GRM matrices may be read in in a DENSE form as well as the previous sparse form. ASReml can now handle singular GRM matrices. More.

Reading Data

Combining columns from separate files

The new MERGE directive is a replacement to the former !MERGE facility.

Datafile line qualifiers

Data folder

A !FOLDER qualifier has been added to facilitate setting path to the folder containing data files.

Factor level combination.

The !GROUP and !FAMILY qualifers allows formation of a new factor defined by grouping levels of an existing data factor. The !FAMILY qualifier takes the redefinition from a file.

Holding variance components fixed.

The !HOLD qualifier can be used to stop ASReml updating certain variance parameters in a temporary manner.

My Basis Function

The !MBF directive has several new options which facilitate its use to read in QTL marker data.

Linear Model Specification

More than 20 dependent variables

The dependent variable can now be a set of variables grouped by the !G factor definition qualifier.

Trait dependent covariates

A vect() model term has been added to allow traits specific covariates to be fitted for !G grouped variables.

Generalized Linear Models

The !MULTINOMIAL qualifier allows the fitting of multiple threshold categorical trait models.

Variance structures.

Direct sum variance structures

The !SUBSECTION qualifier facilitates the definition of a direct sum variance model for example a series of independent time series with common variance and correlation parameters.

Constraining variance parameters.

The manner of setting constraints under the !VCC qualifier has been extended so as to make it easier to specify a large number of simple constraints.

Factor Analytic updates.

The updates of loadings in factor analytic models has been revised in association with the !AILOADINGS qualifier in an attempt to make it easier to get convergence.

PREDICT

General

The format of the predict header has been modified in an attempt to make it easier to understand.

Associateded factors.

Several users have had difficulty getting sensible predictions when they have large factors which have a hierachical or nested, treelike association among the levels, for example 1000 genetic lines in 30 families. The !ASSOCIATE qualifier facilitates defining these relationships for the PREDICT statement and for multilevel nesting, averaging in various ways.

Calculating derived Variance parameters.

The .pin file postprocessing option has been linked into the main .as file processing with the VPREDICT statement.

Concerning residuals.

A new procedure has been implemented for examining the size of random effects, including the residuals. It is invoked by the !OUTLIER qualifier.

A new file type has been added when residuals are saved to a binary file. The .vll file holds level labels so that this information can be assiciated with the factor variables in the binary file if the data is read back for further processing.

Experimental options.

Imputation

The !IMPUTE qualifier invokes an iterative process of fitting a large model by cycling between two simpler submodels.

Singular G

The !SING qualifier invokes an experimental method of fitting singlar variance matrices through a Cholesky factorization.

Return to index