XXI. Summer School of Biometrics

June 26–30, 2017, Karlov pod Pradědem, Czech Republic


Lectures of the Preparatory Workshop

Biometrical Approaches in Genetics and Breeding of Farm Animals
Josef Přibyl, Jana Přibylová, Luboš Vostrý

Genetic work with farm animals is organized on a state level and is regulated by low. Farmers are organized in Association, which are responsible for breeding programs. Breeds are continuously selected and improved to the demands of farmers. Base for selection is regular production recording in herds, which covers approx. 95% of dairy cows (350,000 cows), 10% of beef breeds cows (22,000 cows), 10% of sheep (25,000 ewes) and 5% of pigs (5,000 sows) and their progeny. Data are centralized a stored several decades. Data included production, evidence of animal and farm and pedigree. Data are regularly several times in a year BLUP evaluated by authorized organization (Plemdat) and distributed to farmers. Procedures of evaluation are regularly internationally validated. In dairy cattle are results – estimated breeding value (EBV) of sires from participating countries delivered to the global evaluation in Interbull, because many sires have progeny in several countries. A global evaluation is by MACE method and to the country is returned new EBVs of all globally used sires, converted into national scale of each country. Research is under mentioned conditions. The methodology is based on analysis of surveys. Basic procedure is Mixed Linear Model. Files cover several hundred thousand till several tens of millions of observations. Evaluation is in several steps. Editing – elimination of unreliable values and classes with small number of observations. Creation of effects and pedigree files. Tuning of models with fixed effects (LSM). Construction of mixed model. Calculation of variance components (REML, Gibbs Sampling) on a fitted data. Implanting of variance components into BLUP for evaluation of all animals. Different models for BLUP are used, ST/MT, Animal, Maternal.

Elements of Experimental Design in Plant Variety Testing
David Hampel, Jiří Hartmann, Martin Tláskal

Realization of an experiment is affected less or more by external conditions. In the case of plant variety testing in one growing cycle, the main influence is often identified as a soil heterogeneity. To avoid this, we should apply not only correct agrotechnics, but also appropriate design of an experiment. This presentation contains the basic ideas of experimental design. Concepts of randomization, complete and incomplete blocks, etc. are discussed. With Alpha design as an example, the criteria of design efficiency are presented. Different designs are presented briefly. Moreover, problems related to experimental series design are discussed as well.

Basic Models in Plant Variety Testing
David Hampel, Jiří Hartmann, Martin Tláskal

The main purpose of a plant variety experiment realization is a comparison of particular plant varieties from different insights: yield, frost resistance, length, content of chemical substances. To achieve this, it is necessary to use appropriate models and relevant techniques of the statistical inference. In this presentation, a complete procedure of one-factorial experiment evaluation is presented, including assumptions testing (normality, variance homogeneity, independence), general hypothesis building, pairwise comparison, links to regression and analysis of variance. Moreover, extension for two- and more factorial models is given and the related term “interaction” is explained. Further, mixed models are introduced briefly; estimation using REML method is described as well. Examples of problems leading to different types of models are shown, for example nonparametric approach, generalized linear model for binomial data, nonlinear joint-regression model.

Elements of Selected Multivariate Methods
David Hampel, Lenka Viskotová

When dealing with data, where objects are described by more than one characteristics, researchers face serious problems at different levels: data visualization, multicollinearity problem, characteristics with weak information ability, etc. There are methods constructed especially for this type of data. Firstly we will deal with hierarchical cluster analysis, where we can measure similarities between objects and establish clusters of similar ones. The type of distance between the clusters plays an important role here. Further we investigate Principal components analysis, where the set of artificial independent variables is constructed based on original characteristics. The percentage of explained variability is observed and real dimension of the exercise is visible. Based on these results, Factor analysis can be performed to find several factors describing the original dataset with a negligible loss of information. These artificial factors should be identified using correlation to the original variables (so-called factors rotation is often used to improve this ability). Graphs based on interim results of factor analysis can be used to visualize multidimensional data relationships