Articles
Flexible linear mixed models for complex data in horticultural tree breeding
Article number
1362_19
Pages
139 – 146
Language
English
Abstract
Response to selection and improved confidence in the value of new germplasm from breeding is dependent on objective and accurate prediction of genetic effects from available data.
However, data from horticultural tree crop breeding may be complex.
Here we review opportunities for implementing linear mixed models (LMM) for several examples of complex data in horticultural tree crop breeding to obtain the best (most accurate) and unbiased predictions (BLUPs) of the genetic effect of candidates.
Firstly, LMM theory for a simple field trial is presented.
We then extend LMMs for multiple random observations within an experimental unit (level at which genetic treatment is applied) and demonstrate that multiple sampling from a single tree cannot be considered replication of genetic effect and will lead to incorrect estimates of heritability.
Many traits exhibit a trend with tree age, and LMMs are presented that allow for heterogenous genetic and residual variances, and correlations, among repeated measures across experimental units.
Thirdly, LMMs that account for within trial spatial variation and spatio-temporal correlation are presented.
Fourthly, the flexibility of LMM to model complex data from multi-environment trials and to evaluate genotype-by-environment interaction is discussed, and extended to multi-trial, season within trial, and trait data.
Lastly, we describe the use of LMM for data generated from complex multi-tier designs.
The examples discussed here demonstrate the flexibility of LMMs modelling complex data in horticultural tree crops breeding and biases that can occur with incorrect or simpler models.
However, data from horticultural tree crop breeding may be complex.
Here we review opportunities for implementing linear mixed models (LMM) for several examples of complex data in horticultural tree crop breeding to obtain the best (most accurate) and unbiased predictions (BLUPs) of the genetic effect of candidates.
Firstly, LMM theory for a simple field trial is presented.
We then extend LMMs for multiple random observations within an experimental unit (level at which genetic treatment is applied) and demonstrate that multiple sampling from a single tree cannot be considered replication of genetic effect and will lead to incorrect estimates of heritability.
Many traits exhibit a trend with tree age, and LMMs are presented that allow for heterogenous genetic and residual variances, and correlations, among repeated measures across experimental units.
Thirdly, LMMs that account for within trial spatial variation and spatio-temporal correlation are presented.
Fourthly, the flexibility of LMM to model complex data from multi-environment trials and to evaluate genotype-by-environment interaction is discussed, and extended to multi-trial, season within trial, and trait data.
Lastly, we describe the use of LMM for data generated from complex multi-tier designs.
The examples discussed here demonstrate the flexibility of LMMs modelling complex data in horticultural tree crops breeding and biases that can occur with incorrect or simpler models.
Authors
C. Hardner, J. De Faveri
Keywords
unbalanced data, repeatability model, pseudo-replication, repeated measures, spatial analysis, genotype-by-environment interaction
Groups involved
- Division Plant Genetic Resources, Breeding and Biotechnology
- Division Ornamental Plants
- Division Tropical and Subtropical Fruit and Nuts
- Division Vegetables, Roots and Tubers
- Division Temperate Tree Nuts
- Division Temperate Tree Fruits
- Division Vine and Berry Fruits
- Division Greenhouse and Indoor Production Horticulture
- Division Postharvest and Quality Assurance
- Division Horticulture for Human Health
- Commission Agroecology and Organic Farming Systems
- Working Group Genetic Transformation and Gene Editing
Online Articles (85)