Curriculum Vitae
Arthur R Gilmour,
Biometrician,
My recognition as a
Principal Research Scientist (Biometrics) is primarily based on my part in the
development of ASReml (A S Rem~el). ASReml is a major statistical computing
program developed since 1996 and currently used under commercial license in at
least 15 countries especially in the disciplines of plant, tree, animal and
fish breeding. I have written virtually
all the 88000 lines of Fortran code in ASReml. While, I rely heavily on my principal
colleagues Brian Cullis and Robin Thompson for theoretical
rigor, I am not just a programmer as this document should reveal.
I joined NSW Agriculture as a trainee
in 1967 while studying Agricultural Science at
My research focus is statistical computing in the area of linear models. Three major programs arising from this research are REG (1973-1993), BVEST (1992-1998) and ASREML (1996-present). BVEST is still used by the DPI Advanced (Sheep) Breeding Services although no longer used for LAMBPLAN processing. These programs (and many smaller programs written over the years) arose from needs of fellow researchers to undertake analyses not otherwise available. That is, the theory was available (in general if not specifically) but no software was available to facilitate its application to production data.
My role as a biometrician includes providing statistical designs, consulting with researchers, analysis of data and assistance with reporting. This continuing role provides the stimulus for the software developments. That is, my research is ‘data driven’, a response to particular problems encountered directly in my consulting, or by my colleagues, to which I provide a general solution rather than just solving the immediate problem. I also have a significant role mentoring in statistics and its application to sheep breeding research.
I retired from NSW DPI in January 2009 after 41 years service and am now free to consult privately with breeders and other users of ASReml and BVEST. I will continue to support ASReml.
First I refer to the 22 papers that have been published between
2005 and 2008.
This publication list shows I am still active as a senior
biometrician mentoring and directing important production research. I am consulted when there is some particular
complex issue with the analysis.
There is one paper (96) on QTL analysis of which I am sole
author with 2 related conference papers (C63 and C66). The paper describes a mixed model approach to
QTL detection in F2 or backcross populations which I developed and have
implemented in ASReml. Simulation
studies show that it is more effective than other popular methods in
identifying regions where a QTL appears to occur based on mapped marker
data. As a mixed model, it can
effectively adjust for confounding factors in the data (such as spatial
variation) and additive genetic relationships.
The conference papers describe its successful application to a cattle
breeding mapping experiment.
My major research is associated with ASReml (B45, B46). It includes developing and maintaining the
program, training others in its use (W7, W8, W9, W10, W11, Audio tutorials),
supporting users from around the world principally by email and to an
inadequate degree, documenting the procedures used (96, C60, C61, C62).
ASReml 1 was commercially released in 2003, ASReml 2 was
released in 2006 and the release of ASReml 3 is anticipated in July 2008. I will briefly recap the background to ASREML
before describing the recent developments.
The theory behind Restricted Maximum Likelihood (REML) was
published by Patterson and Thompson in 1971.
I met Robin Thompson in 1981 in
The decision by NSW DPI to commercialize ASReml is two
edged. VSN International was contracted
to be the commercial partner. Under the
Contract, NSW has 51% of IP and Rothamsted Research
has 49% of IP. VSN is a venture company
set up my Rothamsted and NAG principally to
distribute Genstat, a major statistical software
system developed at Rothamsted and formerly distributed
by NAG. ASReml is the only other product
VSN distribute. Note that Genstat actually uses the core of ASReml for its REML
processing. Under the contract, NSW DPI
receives one third of sales revenue. NSWDPI received $63,736 in royalties from
sales in 2007. The advantage of
commercialization is that we know ASReml is currently licensed in over 15
countries by over 170 individuals and organizations. Most of these are multiple user
licenses. Over 280 individuals are
currently registered on the ASReml discussion list. There are an unknown number of users with
licensed copies of ASReml 1 (who have not updated) and early versions (pre
2002) which did not require a license.
David Butler of QDPI has developed an S language version of
ASReml for the R and S statistical environments. R is a free (public domain) reimplementation
of the S language. Both provide comprehensive statistical analysis environments
with a huge number of statistical methods including some procedures for mixed
models. ASReml-R has been formally
distributed by VSN over the last few years and is much more comprehensive than
the other mixed model procedures otherwise available. However, R users are generally resistant to
paying for software and the R developers will not promote procedures like
ASReml-R that require a license.
Since commercialization, ASReml has not been available to
third world users (without
western funds). I am
hopeful that this restriction will be removed with the release of a ‘Discovery’
version when ASReml 3 is released. From
my perspective, income from sales is less important than facilitating efficient
analysis, wherever the need.
ASReml has had a huge direct impact in many areas of
Agricultural Research, especially plant and animal breeding. This impact was recognized early in relation
to animal breeding when I was appointed a Fellow of the Association for the
Advancement of Animal Breeding and Genetics at its meetings in
Without deprecating the contribution of others, and especially my colleagues Brian Cullis (SPRS) and Robin Thompson (former head of statistics at Rothamsted), I see my role has been to make new ideas in the area of mixed models accessible to general researchers so they can effectively explore their data. The novel contribution I made was to achieve economies in computation which meant models could be fitted to large bodies of data which previously were too large or complex to be handled.
However, it is in plant breeding that ASReml has facilitated
a paradigm shift, not only to the application of new statistical models, but
too their development. My close
association with Brian Cullis has made it possible
for him to revolutionize the statistical analysis of plant breeding data. Thus, in the last 20 years, broadacre plant breeding in particular has moved from
simple averaging of results from diverse trials to integrated analysis
incorporating spatial adjustment and appropriate trial weighting based on
individual trial variability. Brian has
overseen the development of new models and their deployment across
The major enhancements I have implemented in ASReml 2 and 3 are as follows:
· Testing of fixed effects is not simple in mixed models. I have implemented 2 methods for calculating the denominator degrees of freedom making significance testing of fixed effects possible
· QTL methods are implemented from two perspectives. I have implemented several traditional methods as well as my new method (96). More recently, users have thousands of markers to explore and methods have been implemented to handle this.
· Pedigree extensions: Animal breeders have traditionally used additive genetic relationships based on the pedigree and a diploid genetic structure. I have recently extended this to handle varying degrees of inbreeding typically found in plant pedigrees.
· Generalized linear mixed models was the subject of my PhD (1983) and I have implemented these basic forms in ASReml. More recently I have implemented multiple threshold models for ordered multinomial data.
· Tabulation and various other data exploration procedures have been added to help users explore/verify the structure of their data prior to analysis.
· Prediction of means and trends has proved a challenge for some of the more complex models that can be fitted in ASReml. Recent work has facilitated prediction in hierarchal models.
· Merge is a facility to combine data from several files for integrated analysis.
· Model extensions allow for more sophisticated testing of effects and variances.
ASReml3 consists of some 88000 lines of Fortran
code, nearly all of which I have written.
Documentation includes a User Guide (400 pages), an online help system
of some 158 HTML pages, and a tutorial series consisting of 4 hours of
audio. I drafted and maintain the User
Guide with significant input from collaborators in formatting and the text of
some sections. I have prepared all of
the HELP system and the audio tutorials.
To summarize, my current involvement with ASREML includes
Maintenance: With any major statistical package,
there is a continual need for maintenance as users find novel ways to use the
package. ASREML provides for a very wide
range of models and users sometimes discover bugs or limitations as they invoke
combinations of options never previously used.
Commercialization: As well as providing
the product, documentation and test examples to VSN, I provide support to users
both directly and as requested by VSN.
Training and consultancies: I periodically
conduct courses for new and experienced ASReml users, present details of
statistical models and methods at national and international conferences, and
undertake minor and major consultancies for commercial users,
Developments: Ideas for developments come from three
primary sources: my own ideas for making ASReml better arising from my own use
of it in consulting and from user applications, requirements arising from needs
of the National Statistics project (Brian Cullis) and
ideas developed by Robin Thompson. I periodically meet with Brian and Robin to
review these ideas to establish priorities and formalize them more. The list is long with more sophisticated
modeling of vatriances being high on the agenda.
One rough
measure of the importance of ASReml is that a Web of Science citation report
identifies 879 citations over 11 years (1996 to 2007). There are 507 other citations of my work
including 130 to my 1997 JABES paper describing single site spatial analysis.
While I have little line management responsibility, I continue to provide support to several PhD students, to review papers in several statistical journals and mentor colleagues and junior scientists.
Student Supervision: I am not formally supervising any
students at present but have had significant input into the PhD studies of
Helena Oakey (
Peer review:
I have refereed papers for Australian Journal of Agricultural Research,
Biometrics, Journal of Agricultural Science Cambridge, Genetics Selection and
Evolution, Journal of Agricultural, Biological and Environmental Science, Crop
Science, Journal of Animal Breeding and Communications in Statistics.
Research leadership:
I continue to play a significant mentoring role among my biometrical
colleagues being involved in giving statistical guidance and support, internal
review of draft publications, monitoring the internal consulting report series
and advising on software and hardware upgrades.
I do not have any formal line management responsibilities.
External funding:
I had a major role in the GRDC project Statistical support for plant improvement programs just completed
and have a major commitment to Integrated genetic
parameter estimation of sheep production traits with the Australian Sheep
Industry CRC.
Conferences: I presented an
invited paper (C60, C61, C62, C63, C64) to the 8th
World Congress on Genetics Applied to Livestock Production, held in
I was awarded a plaque by Meat and Livestock Australia in 1995 in recognition of the development of BVEST which underpinned the development of LAMBPLAN.
I was appointed a Fellow of the Australasian Association for Animal Breeding in 2001 in recognition of the impact of ASReml in facilitating the analysis of animal breeding data.
I do not expect to cease involvement in ASReml. Three persons, Simon Harding of VSN, David Butler of QDPI and Damian Collins of NSW DPI have started collaborating to rewrite the core of ASReml with a view to better documenting its procedures, implementing more modern programming techniques and improving algorithms where possible. The goal is that the new core will better handle the new developments required. I am integral to that process as the aim is to document and transfer my knowledge and skill while I am available. I expect I will continue to work on the ASReml program, incorporating the new core in due course, but also implementing more of the developments that have been proposed.
The most interesting idea for future development is the development of hyper models for the variance parameters.
Research and
Publication:
My research represented in ASReml has made and
continues to have a huge impact in both the efficiency and soundness of
statistical analysis with linear mixed models in NSW DPI,
I have maintained a steady publication rate (Figure 1) with 22 papers published since my last interview.
Transfer of Skills:
I constantly support biometricians, statisticians and research officers
from both within and outside NSW Agriculture and
International
Standing: I have very high standing internationally, particularly in
the animal breeding discipline and among other users of ASReml, people needing
to perform complex statistical analysis by linear mixed models.
The following
publications appeared since 2005 (the two paper with * were listed previously
as in press).
Papers
(83)
Fischer,
T.M., van der Werf,
J.H.J, Banks, R.G., Ball, A.J. and
Gilmour, A.R. (2005) Genetic analysis of weight, fat and muscle depth in
growing lambs using random regression models. Animal Science 82:
13-22.
(84)
*
N.M. Fogarty, V.M. Ingham, A.R. Gilmour,
L. P. Cummins, G.M. Gaunt, J. Stafford, J.E. Hocking Edwards and
R.G. Banks (2005) Genetic evaluation of crossbred lamb production. 1. Breed and
fixed effects for birth and weaning weight of first cross lambs, gestation
length and reproduction of base ewes. Australian
Journal of Agricultural Research 56:
443-453.
(85)
*
N.M. Fogarty, V.M. Ingham, A.R. Gilmour,
L. P. Cummins, G.M. Gaunt, J. Stafford, J.E. Hocking Edwards and
R.G. Banks (2005) Genetic evaluation of crossbred lamb production. 2. Breed and
fixed effects for post-weaning growth, carcase and wool of first cross lambs. Australian Journal of Agricultural Research
56: 455-463.
(86)
Safari,
E.,
(87)
Dutkowski, G.W., Costa e Silva, J., Gilmour, A.R., Wellendorf, H. and Aguiar, A. (2006) Spatial analysis enhances modelling of a
wide variety of traits in forest genetic trials
Canadian Journal of Forest Research 36 :
(88)
(89)
(90)
(91)
Morgan,
J, N.M. Fogarty, S.M. Nielsen, A.R.
Gilmour, (2006) Milk yield and milk composition from grazing primiparous
crossbred ewes. Australian Journal of
Agricultural Research 57: 377-387.
(92)
Safari
E, Fogarty NM, Gilmour AR, Atkins KD, Mortimer SI, Swan AA, Brien
F, Greeff JC, van der Werf,
JHJ (2007) Across population genetic parameters for wool, growth and reproduction
in Australian Merino sheep 1. Data structure and non-genetic effects. Australian Journal
of Agricultural Research. 58, 169–175.
(93)
Safari
E, Fogarty NM, Gilmour AR, Atkins KD, Mortimer SI, Swan AA, Brien
F, Greeff JC, van der Werf,
JHJ (2007) Across population genetic parameters for wool, growth and
reproduction in Australian Merino sheep 2. Estimates of heritability and
variance components. Australian Journal of Agricultural Research. 58, 177–184.
(94)
Morgan,
J,
(95)
Eady, S.J., Garreau, H.S., and Gilmour, A.R. (2007) Heritability of resistance to bacterial
infection in meat rabbits. Livestock Science, 112: 90-98.
(96)
Gilmour, A.R. (2007) Mixed model
regression mapping for QTL detection in experimental crosses. Computational
Statistics and Data Analysis 51:3749-3764
(97)
Hogue, M.A., Arthur, P.F., Hiramoto, K., Gilmour,
A.R., and Oikawa, T. (2007) Variance components due to direct
genetic, maternal genetic and permanent environmental effect for growth and
feed efficiency traits in young male Japanese Black cattle. Journal
of Animal Breeding and Genetics. 124:
102-107.
(98)
Rejected
(99)
Safari,
E., Fogarty, N. M., Gilmour, A. R.,
Atkins, K. D., Mortimer, S. I., Swan, A. A., Brien,
F., Greeff, J. C., and van der Werf,
J. H. J.. (2007) Genetic correlations among and between wool, growth and
reproduction traits in Merino sheep. Journal
of Animal Breeding and Genetics. 124:65-72.
(100)
Afolayan, R. A., N. M. Fogarty, N. M., Ingham, V. M., Gilmour, A. R., Gaunt, G. M., Cummins, L. J., and Pollard, T.. (2007)
Genetic evaluation of crossbred lamb production. 3. Growth and carcass
performance of second cross lambs. Australian
Journal of Agricultural Research 58:
457-466.
(101)
Ingham,
V.M., Fogarty, N.M., Gilmour, A.R., Afolayan, R.A., Cummins, L.J., Gaunt, G.M., Stafford, J.,
and Hocking Edwards, J.E.
(2007) Genetic evaluation of crossbred
lamb production. 4. Genetic parameters for first cross animal performance. Australian
Journal of Agricultural Research 58:
839-846.
(102)
N.
M. Fogarty, N. M., Ingham, V. M., Gilmour,
A. R., Afolayan, R. A., Cummins, L. J., Hocking
Edwards, J. E. and Gaunt, G. M. (2007) Genetic evaluation of crossbred lamb
production. 5. Age of puberty and lambing performance of yearling crossbred
ewes. Australian Journal of Agricultural
Research 58: 928-934.
(103)
R.A.
Afolayan, N.M. Fogarty, A.R. Gilmour, V.M. Ingham, G.M. Gaunt and L.J. Cummins (2008)
Genetic correlations between reproduction traits and growth and wool production
in crossbred ewes. Australian Journal of
Experimental Agriculture 48: 8
(104)
Afolayan RA,
(105)
Prayaga KC. Henshall JM. Swain DL. and Gilmour A.R. (2008) Estimation of
maternal variance components considering cow-calf contacts under extensive
pastoral systems Journal of Animal
Science. 86(5):1081-1088
Reference
Manuals
(B45) Gilmour,
A.R., Cullis, B., Gogel,
B.J., Welham, S. and Thompson, R.(2006) ASReml User Guide. Release
2. VSN International,
(B46)
(C57) Safari, E., Fogarty, N.M., Gilmour, A.R., Atkins, K.D., Mortimer,
S.I., Swan, A.A., Brien, F., Greeff,
J.C. and van der Werf,
J.H.J. (2005) Preliminary genetic parameters for clean fleece weight, fibre
diameter, hogget weight and number of lambs born in Merinos. Proceedings
of the Association for the Advancement of Animal Breeding and Genetics, 16: 180-183
(C58) Worsley, P.M., Dominiak,
B.C., Gilmour, A.R. and James, D.G.
(2005) Pilot study of Intra-town dynamics of Queensland Fruit Fly (Bactrocera tryoni). Proceedings of SSC 2005 Spatial Intelligence, Innovation and Praxis:
The national biennial Conference of the Spatial Sciences Institute, September
2005.
(C 59) Fogarty, NM, Safari, E, Gilmour,
A.R., Ingham, VI, Atkins, KD, Mortimer, SI, Swan, AA, Brien,
F, Greeff, JC, van der Werf,
JHJ (2006) Wool and meat genetics – the joint possibilities. In: PB Cronje and D Maxwell (eds.) Wool Meets Meat – Tools for a
modern sheep enterprise. Proceedings of the 2006 Australian
Sheep Industry
(C 60) Gilmour, A.R., (2006). Statistical
models for multidimensional (longitudinal/spatial) data. Proceedings of the 8th World Congress on Genetics Applied to
Livestock Production,
(C 61) Gilmour, A.R.,
(C 62) Gilmour, A.R., and
Thompson, R. (2006). Equation ordering for Average
Information REML. Proceedings of
the 8th World Congress on Genetics Applied to Livestock Production,
(C 63) Esmailizadeh Koshkoih, A. Pitchford W.S., Bottema C.D.K., Verbyla, A.P.,
and Gilmour, A.R. (2006).
MAPPING MULTIPLE QTL FOR BIRTH WEIGHT IN CATTLE USING A MIXED MODEL APPROACH.. Proceedings of the
8th World Congress on Genetics Applied to Livestock Production,
(C 64) Cloete, S.W.P., van Wyk, J.B., Scholtz, A.J., and Gilmour,
A.R., (2006). GENETIC PARAMETERS
(C 65) Safari, E.,
(C 66) Pitchford W.S., Esmailizadeh Koshkoih, A., and Gilmour, A.R., (2007) Combining information
across traits using a factor analytic model increases the power of QTL
detection. Proceedings
of the Association for the Advancement of Animal Breeding and Genetics, 17:
368-375.
(C 67) Afolayan, R.A., Gilmour, A.R., and
ASREML
WORKSHOP/CONSULTANCIES
(W7) Pioneer,
(W8) VSN,
Hemel Hemsted 2 days, 2006
(W9) CSIRO,
Armidale, 2 days, 2007
(W10)
(W11) Pioneer,
Travel
Reports
(R18) Gilmour, A.R., (2005) Overseas travel report 2005 Pioneer -
(R19) Gilmour, A.R., (2006) Overseas travel report January 2006 Rothamsted Research 9pp
(R20) Gilmour, A.R., (2006) Overseas travel report August 2006
Pioneer -
(R21) Gilmour, A.R., (2007) Overseas travel report 2007 ILRI in
(R22) Gilmour, A.R., (2008) Overseas travel report January 2008
Pioneer -
ASREML
Tutorials
Set
of 16 audio tutorials running for 15 to 20 minutes each.
Table 1 Summary of publications over time
|
Scientific |
Conferences |
|
|
||
YEAR |
Collaborative |
Statistical |
Collaborative |
Statistical |
Other |
Total |
71 |
1 |
|
|
|
1 |
2 |
72 |
|
|
|
|
2 |
2 |
73 |
|
|
1 |
|
3 |
4 |
74 |
2 |
|
|
1 |
2 |
5 |
75 |
|
|
|
|
|
0 |
76 |
1 |
|
|
|
|
1 |
77 |
1 |
|
|
1 |
1 |
3 |
78 |
2 |
|
|
1 |
|
3 |
79 |
8 |
|
|
1 |
|
9 |
80 |
|
|
|
|
|
0 |
81 |
3 |
|
|
|
|
3 |
82 |
|
|
|
1 |
|
1 |
83 |
1 |
|
|
|
2 |
3 |
84 |
|
|
|
|
1 |
1 |
85 |
1 |
2 |
|
|
1 |
4 |
86 |
1 |
|
|
1 |
|
2 |
87 |
|
1 |
|
|
|
1 |
88 |
1 |
|
1 |
1 |
1 |
4 |
89 |
2 |
|
|
|
3 |
5 |
90 |
2 |
|
|
|
1 |
3 |
91 |
1 |
|
|
2 |
1 |
4 |
92 |
3 |
1 |
5 |
1 |
1 |
11 |
93 |
1 |
|
|
1 |
1 |
3 |
94 |
6 |
2 |
1 |
3 |
1 |
13 |
95 |
2 |
2 |
1 |
2 |
3 |
10 |
96 |
|
2 |
|
1 |
3 |
6 |
97 |
|
1 |
3 |
2 |
4 |
10 |
98 |
4 |
|
3 |
6 |
1 |
14 |
99 |
2 |
1 |
3 |
1 |
3 |
11 |
2000 |
0 |
3 |
0 |
0 |
2 |
5 |
2001 |
4 |
1 |
0 |
0 |
3 |
8 |
2002 |
2 |
1 |
1 |
1 |
1 |
6 |
2003 |
2 |
2 |
2 |
1 |
0 |
7 |
2004 |
5 |
3 |
0 |
0 |
0 |
8 |
2005 |
4 |
0 |
5 |
1 |
0 |
10 |
2006 |
3 |
0 |
4 |
2 |
1 |
10 |
2007 |
12 |
1 |
3 |
0 |
1 |
18 |
2008 |
2 |
1 |
1 |
2 |
1 |
7 |