Richard L. Quaas - Professor
PhD from Colorado State U.
Graduate fields: Animal Science, Animal Breeding
Area(s) of interest: statistical and quantitative genetics, animal breeding
Teaching:
Email: rlq1@cornell.edu
Current Research
Dr. Quaas sitting astride 'Red Littlebean' on a recent trip to inner Mongolia.
The modern era of animal breeding in domestic livestock began with dairy cattle. It was a result of the confluence of several factors. DHIA provided a large data-base of objective measurements; artificial insemination produced large numbers of progeny in numerous herds; computers became available to 'crunch' the voluminous data and new statistical techniques, primarily the work of C.R. Henderson at Cornell, were invented to prescribe what the crunching should accomplish. The objective was to extract information from the data that would be useful to livestock breeders in selection programs.
Over the years the procedures have become more sophisticated, the computers more powerful and the data-bases much larger. The Cornell animal breeding group has always been at the forefront of this mixture of statistics, computers and genetics of domestic livestock and continues to be so. Work may be categorized as:
a. more realistic models for describing the data;
b. statistical procedures to prescribe how the data should be analyzed;
c. computing algorithms to carry out the analysis.
The "Animal Model" is the current generic model of choice by animal breeders for many applications. It was devised by Henderson long before it could be applied; it was named by Quaas and Pollak as a contrast to our "Reduced Animal Model" which resulted from showing that re-formulating the model lead to the same "answers" but with fewer computations. This model also incorporated maternal effects which has lead to evaluations of beef cattle not only for the "direct" genetic effects but the maternal genetic effects as well. Most recently this work has been extended to a "Multiple Breed Evaluation." This system uses a model which accounts for breed differences and heterosis. It is scheduled to be applied in 1997 for a joint evaluation of American & Canadian Simmental cattle.
Other research has included procedures for dealing with data which are not "nice." E.g., data on length of herd life where the problem is how to account for the length of life of a cow still in production and data on calving ease - discrete subjective scores. In the former case we have adapted "failure time" techniques; in the latter we have extensively studied the so-called "Threshold Model" and currently use it for calving ease genetic evaluations for Simmentals. In both cases the procedures are "non-linear" and required solving new computational demands.
In many animal breeding problems, causal components of variation are needed to either assess the importance of various genetic and environmental factors or to properly assign relative weight to various sources of information, e.g., an animal's own record vs those of it's relatives. Estimating these "variance components" is a demanding task especially in non-traditional models. We are quite active in this area, examples being autoregressive models used in the Test Day Model and in analyses attempting to quantify the impact of various health disorders on milk production. For many of these non-traditional problems, Monte Carlo methods are proving quite useful.
Work in the molecular genetics of domestic livestock is just beginning. Its promise for genetic improvement is great but exactly how is as yet unclear. It is clear, however, that the techniques employed will generate new types of data. Simulation studies have given some insight into how the information might be used in genetic improvement programs. Methods for incorporating such information are being developed in collaboration with workers at Wageningen.