SOLAR Version 6 introduces EVD (Eigenvalue
Decomposition) a method for greatly improving the speed of model
maximization, the central operation of all analysis performed by
SOLAR. Beta version 6.3.0 uses only a simplified version of EVD
called EVD1 which applies only to univariate models and does not
yet achieve the full speedup possible. Nevertheless, we have seen
performance improvements up to 80 times with actual production
data (though the very large pedigree set models we most often use
improve about 3 to 4 fold). EVD is not used for all models, only
when option modeltype evd
is selected or
certain new commands are used.
Following on from earlier SNP association analysis
techniques QTN and QTLD introduced in earlier
versions of SOLAR, a new technique called Measured
Genotype Association Analysis is introduced with the new
command mgassoc
. By
default, mgassoc
sets up and uses EVD1 for the
greatest possible performance. You need only specify the basic
model trait and covariates and mgassoc
takes care of
the rest. When EVD2 or faster techniques are completed, they
will be incorporated in mgassoc
if possible.
An often requested convenience feature has been added. SOLAR now has command line editing, in which previous solar commands can be recalled with the up and down arrow keys, and the cursor can be repositioned within the current command with the right and left arrow keys. This feature is made possible by the free software program rlwrap by Hans Lub which is included in this distribution. The entire distribution is included along with the pre-installed and scripted program. For help with installation issues on your system, should any arise, or for more information about rlwrap, see the system-specific README file for your system, such as README.linux for linux systems.
As usual, there are many additional improvements which may be
viewed with the change-notes
command or seen
in the Appendix 4 Change Notes.
Version 5 was an internal experimental version exploring other
less successful methods for improving maximization performance.
SOLAR Version 4 introduces multivariate analysis of models with up to 20 quantitative and discrete traits, including mixed quantitative and discrete traits. There is nothing special you need to do for multivariate models except specify all the required traits. For example:
solar> trait q1 q2 q3 d4
SOLAR handles all the other details. You can simply use all
the other commands as you would have done otherwise. (See
Chapter 3 for a basic
tutorial.) Note however that that the
-screen
option for
polygenic
is not allowed for multiple
traits, as we believe covariates should be screened for
univariate models. For convenience, all LOD scores computed
by SOLAR are (by default) "1 df effective" LOD scores
which are comparable to those for univariate models. If you
wish to see exactly how a LOD score was calculated give the
command
lod -v
when a linkage model is loaded such as after
running multipoint
. The
lodp
(lod preferences) command can be used to
change how LODs are computed. Further discussion of
multivariate analysis is in Chapter
9.
Another important addition to Version 4 is the
define
command, which can be used to define any
variable expression for use as a trait or covariate. A
useful inverse normal transformation is supplied for use
in the define command, and can also be applied to data
in any file using the
inormal
command.
Association analysis for SNPs can be done using the
qtld
command after preparing the SNP data using the
snp qtld
command. Other useful analyses can be performed
by other options of the snp
command
and the snphap
command.
For a complete list of changes, see the Change Notes.