Students’, colleagues’ and research partners’ experience about work and accomplishments from collaborating with Robin Thompson

Robin supervised my PhD (1986–1989) together with Bill Hill. I remember that my first supervisors’ meeting was terrible, because I did not bring ideas to the table, expecting to be told what to do. I soon learned how to manage my supervisors—an important life skill recommended for all prospective PhD students! I found if I led those meetings, showing that I had been working and thinking that I was presented with a tsunami of suggestions on how to move forward. Robin's supervision particularly focussed on prediction of inbreeding in the context of selection (Wray & Thompson, 1990; Wray, Woolliams, & Thompson, 1990 and Wray, Woolliams, & Thompson, 1994), a problem that had perplexed the field but had remained unresolved following Alan Robertson's, 1961 paper, which cryptically had understood the key issue (increased relatedness of those selected as parents compared with randomly selected parents), but did not quite follow through into an accurate predictor. I learnt many key approaches to theoretical research, for example, challenging equations by focussing on the parameter boundaries (such as zero or infinity)—sounds simple and obvious, but often I find others have not been trained to think this way. While saying something dumb would lead to the characteristic “Awww Nioomee” response, I knew that once Robin was engaged on a problem he was hooked. A seemingly distracted gaze into space would result in a couple of pages of matrix algebra together with a computational algorithm ready for implementation. I probably learned too late that Robin cared little for the detail of getting each i right in an equation (after all, if the gist is presented readers can derive for themselves, can't they?). His care quota was filled with answering the big questions, and in his genuine concern for people as individuals as they progressed through the ups and downs of research and of life. In our first paper on prediction in inbreeding in selected populations, Robin introduced me to a paper from an Australian statistician (Tallis, 1987), which from its title hides a key contribution of deriving variances and covariances conditional on selection. These days, my research focuses on quantitative genetics of human disease, in which the binary nature of a disease trait (diseased or not) has strong parallels with the binary nature of selection (selected or not). I have introduced the Tallis approach several times in recent years to some talented PhD students and postdoctoral fellows (Peyrot, Boomsma, Boomsma, Penninx, & Wray, 2016; Peyrot, Robinson, Robinson, Penninx, & Wray, 2016; van Rheenen, Peyrot, Schork, Lee, & Wray, 2019)—Robin has had impact in many fields! I left Ireland by boat in 1986 to do my PhD in Edinburgh with Charlie Smith on multiple ovulation and embryo transfer (MOET). When I arrived there, Charlie told me he was leaving for Guelph, so I ended up spending 6 months in Canada with him and the rest of time in Edinburgh working with Bill Hill, John King, Brian McGuirk and the person who became my main supervisor, Robin Thompson. Working with Robin was a weird and wonderful experience. He had huge energy, and his brain did not seem to work like that of other people I knew. He would bounce into my office to get an update, ask provocative questions and provide suggestions and then quickly bounce out again, leaving me with many new ideas to investigate further. He encouraged me to think laterally, and, through him, I learned the importance of scientific rigour. What set him apart for me were, however, his human qualities. Although sometimes frustrated with us, he really was totally dedicated to all of his students and he had a good sense of humour. The symposium papers outline Robin's scientific contributions in many aspects of quantitative genetics. Robin's ideas made leaps in the development of new methodologies, and I often felt for Robin, when he was clearly frustrated that I could not bridge the several steps in theory that he took for granted. Robin promoted a comprehensive statistical analysis of data to derive parameter estimates, rather than using the results from several analyses to provide less precise parameter estimates. Robin's presentation style was certainly unique, and what often started by building on several ideas would rapidly develop to lead to a new approach on a methodology or experimental design. It's 20 years since I last worked with Robin, so comments from researchers, who have recently worked with Robin, will be more insightful. I was supervised by Robin for my Master's thesis (on estimation genetic parameters for type traits in Dutch dual purpose cattle) and jointly supervised by him and Bill Hill during my PhD. It was a wonderful time and I keep looking back with fond memories on that period (1988–1991). As Robin himself scribbled in a first edition of A.W.F. Edwards’ book “Likelihood” which he gave me as a present at graduation, they were “unlikely times”! I didn't have much prior training in statistics, and still have to look up probability density functions on Wikipedia rather than memorize them or derive them from first principles. One of my objectives during my PhD was to learn as much as I could from Robin, not just about the principles of statistical methods, but also about his way of thinking about problems (and coming up with solutions). That was tough! Not only is Robin's notation notoriously sloppy, he “thinks” in linear models and geometry, a trait in common with RA Fisher who, because of poor eyesight, developed an ability to visualize statistical problems in geometrical terms. It was hard to tease those thoughts out of Robin's brain in a way that was understandable to a mere mortal like me, but I kept trying, eventually with some success. One important skill that Robin taught me was how to approach solving problems, both in computer coding (which I don't do much of these days) and in theoretical derivations: keep it simple (a first-order approximation will often do, and a linear model nearly always works for practical applications) and try to see what happens at the limit. It's an approach and skill which I try to teach my students and postdocs. Most of my time with Robin was spent on models and implementation of the estimation of (co)variance components using REML, and it is very satisfying to see that REML is now also the method of choice in human genetics. I got to know Robin when he was at ARCUS and appreciated his advice and input on the work of our Masters and PhD students in animal breeding and genetics. The opportunities for collaboration and to seek Robin's help increased on his appointment in 1983 as Head of Statistics at the BBSRC's Animal Breeding Research Organisation (ABRO), based next door to the University's Institute of Animal Genetics (aka Genetics Department) at Kings Buildings. This provided demand for students and their supervisors when analysing genetic data and led to many collaborations. He was a frequent visitor and great helper and collaborator for the many postgrad students, whose mathematical and statistical abilities and experience varied greatly. He was popular but demanding of students, most of whom “survived.” These included Karin Meyer, Raphael Mrode, Naomi Wray and Peter Visscher, and others who have returned to their home countries and are less well known. Several remained in Edinburgh, including Sue Brotherstone and Aviagen geneticists. Of the many papers Robin co-authored with me, for many of which the students did much of the work, one that had me most surprised and interested was that on the probability of non-positive definite (non-PD) covariance matrices in multivariate ANOVA (Hill & Thompson, 1978). We had initially become aware of the problems associated with an non-PD matrix when one of our MSc students—later completing PhD (Thorvaldur Arnason, 1983)—examined alternative selection indices incorporating various subsets of information and obtained silly results. To investigate further, we took the simple one-way balanced case and asked what is the probability of NPD even in this simple design (Hill & Thompson, 1978). The probability increases greatly the more traits, the smaller the group sizes and the numbers of levels. The results were initially alarming (at least to me): the probability approached 1.0 if many traits were involved: most estimates were not in the parameter space—that is, impossible. Increasing numbers of traits exacerbated the problem. Subsequently, we suggested ways of “bending” (Hayes & Hill, 1981) estimates to get them in the parameter space wriggling along the boundary. Our method does not just stop estimates going out of bounds but it is intended to get better information. Nowadays, more formal methods are used (in ASREML and Wombat, e.g., Meyer & Kirkpatrick, 2010) to keep estimates within the boundaries. It is a great pleasure to pen a few words for this issue in honour of Professor Robin Thompson. As others have described, Robin has made huge contributions to the science underpinning animal and crop breeding programmes globally, and much more besides. His many collaborations with the various Edinburgh groups working in this area had, and continue to have, particular impact in UK dairy, beef and sheep breeding programmes, and I will focus on a few of these here. I first met Robin in the early 1980s, when he moved from the AFRC Unit of Statistics to join the AFRC Animal Breeding Research Organisation (ABRO, later part of the Roslin Institute), where I was completing my PhD studies on the genetics of lean growth and efficiency in beef cattle. I made a habit of acquiring additional, informal advisors, and Robin was semi-willingly recruited to one of these roles! Robin's advice was especially helpful in deriving the statistical explanation of why responses to selection on “biological indexes” (e.g., lean growth) were dominated by the most variable component of the index (e.g., growth rate). Our paper on this (Simm, Smith, & Thompson, 1987) is not one of his best-known works, but there is still time! For me, it was the first of many, hugely rewarding, stimulating, productive and very enjoyable collaborations over these 30 years. On completing my PhD, I moved to the East of Scotland College of Agriculture (ESCA; later part of the Scottish Agricultural College (SAC), and now Scotland's Rural College (SRUC)). In ESCA, I assumed responsibility for the Langhill Dairy Cattle Breeding Project, initiated in the 1970s by ESCA, the University of Edinburgh and ABRO, and a newer selection project in Suffolk sheep. Over the next 15 years or so, Robin provided regular advice to me and a growing number of colleagues and students working on these and related projects—spanning biological and economic impact of selection, development of broader breeding goals and selection criteria addressing health, welfare and sustainability, and breeding programme design. Robin, our colleagues and I, also won funding for a number of animal breeding research projects and studentships with livestock breeding sector partners. One of the first of these, in the early 1990s, was to support the introduction of BLUP evaluations for beef breeds in the UK, by the Meat and Livestock Commission (MLC and later its Signet breeding service). We worked on this with a group of ESCA/SAC and Roslin Institute colleagues, including Naomi Wray, David Nicholson, Ron Crump and Robert Findlay, Julian Bryan and his colleagues at MLC, and some visionary beef breeders, including Richard Fuller and Richard Oates (see, e.g., Crump, Simm, et al., 1997; Crump, Wray, Thompson, & Simm, 1997). It was an exciting, rewarding and challenging project. The challenges were both technical and sociological—the latter, as not all breed societies and their members were enthusiastic about the new approach. Also, as breeders and their customers had access to across-herd genetic evaluation results for the first time, breeders’ reputations were lost and made in the process. Robin's statistical expertise was obviously pivotal, but his farming roots also helped to stimulate uptake. I remember the surprise among some breeders when Robin happily got in amongst some bulls to hang up display boards in preparation for an open day—they had labelled him as a “number cruncher” until that time, and he won some new supporters! Research sponsors were enthusiastically embracing new project management methodology at the time, and it has to be said that Robin was not always as enthusiastic about this approach and the challenging milestones set, as we began wrestling with vast amounts of data of variable quality. With BLUP evaluations in place, Peter Amer and Tim Roughsedge joined the team, and we moved onto developing evaluations for new traits, and a new suite of indexes for different sectors of the beef industry (Roughsedge, Amer, Thompson, & Simm, 2005a, 2005b; Roughsedge, Thompson, Villanueva, & Simm, 2001). These provided tools for breeders and clients to better differentiate between terminal sire and maternal breeding goals. With Dick Esslemont and colleagues at the University of Reading, we worked on genetics of health and fertility in dairy cattle (e.g., Kadarmideen, Thompson, & Simm, 2000; Pryce, Veerkamp, Thompson, Hill, & Simm, 1997; Pryce et al., 1998), building on the earlier work at Langhill with Jennie Pryce, Roel Veerkamp, Bill Hill and Sue Brotherstone. This work helped underpin the development of broader UK breeding indexes. We also worked with John Woolliams, Beatriz Villanueva and University of Reading colleagues on the impact of new reproductive technologies on genetic improvement of ruminants—highlighting the trade-offs between faster response and loss of genetic variation. Around the turn of the millennium, UK levy bodies began to consider outsourcing livestock genetic evaluations. Many of us including Robin, Bill Hill and colleagues in the University, and Mike Coffey who had recently joined SAC, felt this was a threat to the “route to market” for much local breeding research. So, we mounted a bid to create a research-led, UK-based evaluation service, which ultimately became EGENES, led by Mike Coffey. This to support the of new evaluation methods, breeding technologies and of data among in the UK livestock breeding As we look back over our it is often the people we have worked with that Robin's technical expertise is very who has worked with him will also have of Robin's and This of that Robin is in both huge and by so many like me, who have had the to work with On of so many of us, it is a great pleasure to be to that here. In the UK dairy breeding programme by the accurate breeding estimates of different traits have been into indices of genetic (which have been labelled and more recently The project by was by its in by the whose Breeding and then into were from the or with very genetic in were then to in It was not to me (and I was with the how to the The was to information with data of and in the The steps were the information to UK these so they are to and then these with the or from within the I the analyses to be in my office at on a computer which was what be a the I to was a & In my of statistical I to Robin for help to the genetic of the various the analyses was not Robin and spent a in the office at and he to that could not be with and suggested that the was for him to be back to the for an earlier to I was to we had not on the from the Robin have the and asked to be back to I quickly made a We then quickly had to on and were to a between Masters and their the on and information could not differentiate the When I was a at Roslin in Robin was working on an experimental design for multivariate genetic parameters & Thompson, work in at that time was focussed on genetic parameters for trait components in We were interested in and genetic amongst ovulation and and to Robin in the project. What we did was probably the first selection Thompson, & Robin to with and and over some we were in a we Robin to be of the and also a Robin (in at the for the and early Robin and Research of the at I that the key was the For example, in one he got so about and in the that he few to his home in his who he also the to have time and for on his most research the work we did was in one way or other focussing on Robin's first PhD with REML on of and the on the way to the of more and more Robin the to be in the matrix a the was a Robin to get and to and Our colleagues and from Research in had started to multivariate in the & Thompson, Our focus was on the that is, the estimates in the parameter space even with Robin suggested a very in case of a us a of to the & Thompson, challenging ideas were of the the of 1990s, we became interested in the of methods for the estimation of the prediction variances in the first of That to be the next PhD by Although the of our was we were also on breeding Robin us a project for the development of computational for the UK model and, later for the development of breeding estimation et al., In some other Robin often when we to parameters from the for example, simple like from or using animal model information. Robin, his with the ASREML development several PhD students and projects to investigate genetic variation. For example, the of models for genetic evaluation models in dairy cattle and growth in other from the introduction of models for genetic several other research and were by the of ASREML and these I my PhD thesis with the of the dairy Hill, & Thompson, especially for and research papers in those early years and several PhD students were in to the genetic over time, between the traits et al., & Veerkamp, Veerkamp, & & Thompson, In the of this there was a of in genetic evaluation of The of ASREML to genetic parameters with a model to the to breeding for with type model & 2001). has in the model in the and research in PhD thesis & Veerkamp, these models were the of and model genetics in of & Veerkamp, & Veerkamp, & Veerkamp, & Veerkamp, It became very during an ASREML in that the models of which ASREML was the of their breeding estimation the Robin was in the to developing a new as a to with the by Robin, it became that a collaboration between the in and University and Research was the to This has in a very of all recent genetic evaluation including those with of The is used by many breeding for their genetic evaluation The of our collaboration to to The University of has recently the which has been made through a to the in support of the for and and its This is for interested in of model theory in and animal breeding and a to in or even The for the is based on the where I and Robin or up I first met Robin when I was a PhD in the the supervision of Roel Robin would Roel who encouraged students to time with We would Robin on our work, and he would and teach us and us to think and teach Robin has always been very and his ideas As a I spent years of When I would for I at He would provide a few at the and we would up on in animal breeding and our I learned so much but always had to be on my Robin is in his science and the of On one my We had a and my still always as to how Robin is This of people of Robin over and and I a symposium in honour of Robin to his of the of a on 30 It is the that the University of Edinburgh can Robin was in for his and in the of animal and breeding and the it was to see person person in about Robin as a collaborator and more people like

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.