Applied Statistics in Animal and Plant Sciences

Joint meeting of Rothamsted Research and the British and Irish Region of the International Biometric Society

(to mark the retirement of Robin Thompson from official duties at Rothamsted)

This meeting is being held on Thursday 19^th January 2006, in the Fowden Conference Hall, Rothamsted Research, Harpenden, Herts., AL5 2JQ. It has been organised by Sue Welham and Frank van den Bosch of Rothamsted Research on behalf of the British and Irish Regional Committee.

Please note that although there is no charge for this meeting, pre-registration is necessary.

If you plan to attend this meeting, please notify Sue Welham before January 10^th 2006.

Session I: (Chair: David Balding (Imperial College))

14.00 Opening Remarks - Joe Perry (President, International Biometric Society British and Irish Region)

14.00 – 14.30 REML, Rothamsted and Robin: a Review

Roger Payne (VSN International)

The methodology of residual maximum likelihood (REML), developed with Desmond Patterson at Edinburgh in 1971, represents the first of Robin Thompson’s major contribution to Statistics. REML provided a reliable and efficient framework for the analysis of linear models that contain several sources of random variation and whose component terms are not balanced (as they would be in ordinary analysis of variance). It also provided a precursor of Robin’s future career at Rothamsted, in that it agreed with the Rothamsted-derived methodology of general balance where both were relevant, and extended the ideas to a wider range of situations.
This talk will review REML and its history, and discuss some of Robin’s related contributions to Rothamsted and its statistics.

14.30 - 15.00 Statistical interactions between Robin Thompson and quantitative genetics

Bill Hill (University of Edinburgh)

Robin Thompson is distinguished for his many fundamental contributions to the study of quantitative genetics. Some high points will be reviewed. Much of this has been through the continuing development and application of REML, which has long been a standard tool in analysis of lab animal and livestock data but is now extensively used in natural and human populations. Through his research and advice he has led the development and application of mixed model methods for genetic evaluation of livestock. Other basic contributions with his many students and colleagues have included analyses of gene flow and population structure.

15.00 – 15.30 Future possibilities in the dissection of quantitative genetic variation

Chris Haley (Roslin Institute)

Utilisation of quantitative genetic variation has always lain at the heart of animal and plant breeding. However, it is only in the last 20 years that efforts to dissect and understand this variation have grown substantially, fuelled by the development of new statistical and molecular technologies. A starting point is the ability to separate genetic from environmental influences in outbred populations, to which the development of REML has contributed mightily. From its stronghold in livestock, REML has established footholds in the study of other natural outbred populations, including humans. Furthermore, when combined with genetic marker information, REML allows the separation of genetic variance associated with the whole genome, with that linked to a specific chromosomal region, i.e. the detection of individual quantitative trait loci (QTL). This provides one of the few tools capable of analysis of data from deep and complex outbred pedigrees in livestock, humans and other species. When using information only from linkage, this approach generally does not map QTL sufficiently precisely to permit their resolution at the molecular level. However, inferring genetic relationships prior to the known pedigree by making use of information on linkage disequilibrium and marker haplotypes can significantly improve mapping resolution. Nonetheless, in many cases mapping of QTL will need to be augmented by other sources of information to resolve the molecular causes of variation. One approach is to use the same tools to explore the genetic control of variation in gene expression and relate this to the functional traits of interest to identify genes that may underlie the QTL. Ultimately, such genome-scale approaches are likely to allow us to understand the genetic basis of a proportion of the variation and hence improve predictions of genetic merit in livestock and disease outcomes in humans.

15.30 – 16.00 Tea break

Session II: (Chair: Chris Rawlings (Rothamsted Research))

16.00 – 16.30 Colonial Collaborations - applications of REML for plant improvement

Brian Cullis, Alison Smith (NSW-DPI), Ari Verbyla (BiometricsSA), Jo Stringer (BSES)

On a colonial visit by Prof. Robin Thompson in the early 1990s the seeds were sown for a collaboration that has, over the course of 15 years, revolutionised the approach to varietal selection in Australian plant improvement programs. It all began with an airline boarding pass etched with a cryptic version of the average information algorithm for REML estimation of variance parameters. This, combined with the need to conduct a variance component analysis for a large unbalanced set of variety by environment data helped give rise to a computer program called AIREML. Of course this has evolved into the package now known as ASReml. ASReml has become the ‘gold standard’ for mixed models software in terms of the complexity of variance models that can be fitted, the ease with which quantities of interest can be predicted and the ability to handle large unbalanced data-sets. These features are underpinned by statistical research, the majority of which has been undertaken as collaborative work with Prof. Thompson. The presentation will summarise some of this work, in particular key research aimed at improving genetic gain in plant improvement programs. The topics include multiplicative mixed models for variety by environment data, competition models and the more general issues of prediction and inference in linear mixed models.

16.30 – 17.00 The challenge of bioinformatics

Chris Glasbey (BioSS)

This is in an amazing, revolutionary time in biology, as the molecular secrets of life are progressively revealed. Bioinformatics is the science of handling and interpreting genomic and post-genomic data to achieve these breakthroughs. Particular challenges include DNA sequence alignment, modelling of gene expression, prediction of protein structure, protein-protein interactions and inferring evolutionary trees. However, the holy grail is a holistic understanding of how biological systems function, which necessitates modelling and combining data from many sources.

17.00 – 18.30 Drinks and snacks.

14.00 – 14.30	REML, Rothamsted and Robin: a Review
Session I:	(Chair: David Balding (Imperial College))
14.00	Opening Remarks - Joe Perry (President, International Biometric Society British and Irish Region)
	Roger Payne (VSN International)
	The methodology of residual maximum likelihood (REML), developed with Desmond Patterson at Edinburgh in 1971, represents the first of Robin Thompson’s major contribution to Statistics. REML provided a reliable and efficient framework for the analysis of linear models that contain several sources of random variation and whose component terms are not balanced (as they would be in ordinary analysis of variance). It also provided a precursor of Robin’s future career at Rothamsted, in that it agreed with the Rothamsted-derived methodology of general balance where both were relevant, and extended the ideas to a wider range of situations. This talk will review REML and its history, and discuss some of Robin’s related contributions to Rothamsted and its statistics.
14.30 - 15.00	Statistical interactions between Robin Thompson and quantitative genetics
	Bill Hill (University of Edinburgh)
	Robin Thompson is distinguished for his many fundamental contributions to the study of quantitative genetics. Some high points will be reviewed. Much of this has been through the continuing development and application of REML, which has long been a standard tool in analysis of lab animal and livestock data but is now extensively used in natural and human populations. Through his research and advice he has led the development and application of mixed model methods for genetic evaluation of livestock. Other basic contributions with his many students and colleagues have included analyses of gene flow and population structure.
15.00 – 15.30	Future possibilities in the dissection of quantitative genetic variation
	Chris Haley (Roslin Institute)
	Utilisation of quantitative genetic variation has always lain at the heart of animal and plant breeding. However, it is only in the last 20 years that efforts to dissect and understand this variation have grown substantially, fuelled by the development of new statistical and molecular technologies. A starting point is the ability to separate genetic from environmental influences in outbred populations, to which the development of REML has contributed mightily. From its stronghold in livestock, REML has established footholds in the study of other natural outbred populations, including humans. Furthermore, when combined with genetic marker information, REML allows the separation of genetic variance associated with the whole genome, with that linked to a specific chromosomal region, i.e. the detection of individual quantitative trait loci (QTL). This provides one of the few tools capable of analysis of data from deep and complex outbred pedigrees in livestock, humans and other species. When using information only from linkage, this approach generally does not map QTL sufficiently precisely to permit their resolution at the molecular level. However, inferring genetic relationships prior to the known pedigree by making use of information on linkage disequilibrium and marker haplotypes can significantly improve mapping resolution. Nonetheless, in many cases mapping of QTL will need to be augmented by other sources of information to resolve the molecular causes of variation. One approach is to use the same tools to explore the genetic control of variation in gene expression and relate this to the functional traits of interest to identify genes that may underlie the QTL. Ultimately, such genome-scale approaches are likely to allow us to understand the genetic basis of a proportion of the variation and hence improve predictions of genetic merit in livestock and disease outcomes in humans.
15.30 – 16.00	Tea break
Session II:	(Chair: Chris Rawlings (Rothamsted Research))
16.00 – 16.30	Colonial Collaborations - applications of REML for plant improvement
	Brian Cullis, Alison Smith (NSW-DPI), Ari Verbyla (BiometricsSA), Jo Stringer (BSES)
	On a colonial visit by Prof. Robin Thompson in the early 1990s the seeds were sown for a collaboration that has, over the course of 15 years, revolutionised the approach to varietal selection in Australian plant improvement programs. It all began with an airline boarding pass etched with a cryptic version of the average information algorithm for REML estimation of variance parameters. This, combined with the need to conduct a variance component analysis for a large unbalanced set of variety by environment data helped give rise to a computer program called AIREML. Of course this has evolved into the package now known as ASReml. ASReml has become the ‘gold standard’ for mixed models software in terms of the complexity of variance models that can be fitted, the ease with which quantities of interest can be predicted and the ability to handle large unbalanced data-sets. These features are underpinned by statistical research, the majority of which has been undertaken as collaborative work with Prof. Thompson. The presentation will summarise some of this work, in particular key research aimed at improving genetic gain in plant improvement programs. The topics include multiplicative mixed models for variety by environment data, competition models and the more general issues of prediction and inference in linear mixed models.
16.30 – 17.00	The challenge of bioinformatics
	Chris Glasbey (BioSS)
	This is in an amazing, revolutionary time in biology, as the molecular secrets of life are progressively revealed. Bioinformatics is the science of handling and interpreting genomic and post-genomic data to achieve these breakthroughs. Particular challenges include DNA sequence alignment, modelling of gene expression, prediction of protein structure, protein-protein interactions and inferring evolutionary trees. However, the holy grail is a holistic understanding of how biological systems function, which necessitates modelling and combining data from many sources.
17.00 – 18.30	Drinks and snacks.