INVESTIGATING EFFICIENCY OF SELECTION USING UNIVARIATE AND MULTIVARIATE BEST LINEAR UNBIASED PREDICTORS. Richard Kerr Tony McRae

Transcription

1 INVESTIGATING EFFICIENCY OF SELECTION USING UNIVARIATE AND MULTIVARIATE BEST LINEAR UNBIASED PREDICTORS Richard Kerr Tony McRae

2 . demonstration of the limited advantage in using multivariate analysis I forgot to mention!!! under scenarios where all individuals are measured for each trait

3 Structure of the talk Summarise the STBA approach to genetic value prediction Summarise the STBA Pinus radiata breeding population In terms of the numbers of families, genotypes, parents etc In terms of the numbers of measurements Then briefly describe a simulation which resembles the radiata program in Australia Why? In order to gauge how much additional gain has been realised from using multi-trait analysis Results of the simulation Conclusion

4 Genetic improvement - you start with an objective Genetic value prediction is complicated, so you need a structure Profit Weights Economic Trait BV Correlations Measured Trait BV $250 per m 3 /ha $600 per Gpa $ MAI STIFFNESS SWEEP 0.5 Early Growth Late Growth 0.4 DENSITY 0.9 -$100 per mm/m STEMST MOE -$400 per cm BRANCH -0.3 BRA BRS -0.9

5 TREEPLAN philosophy is to maximise gain by using all of the data Relationships between measurements Pedigree Genetic correlations Design features and error correlations Trial information Design features Different heritabilities G X E is accommodated by mapping measured traits to selection criteria (SC)

6 GROWTH Variance TAS_0-5 TAS 0-5 WA 0-5 TAS 6-12 WA 6-12 BRA MOE STEMST Define 4 types of correlations WA_0-5 Inter-trait correlation TAS_6-12 WA_6-12 Inter-age correlation BRA Inter-site correlation MOE STEMST Inter-site, inter-age correlation

7 Matrix of correlations among 27 selection criteria P. radiata GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH DENSITY DENSITY DOTHI STEM CGIPP CGIPP CGIPP CVIC CVIC CVIC GTR GTR GTR TAS TAS TAS WA WA WA MVAL MVAL DOTHI DOTHI ST BRA BRQ BRS MOE GROWTH CGIPP GROWTH CGIPP GROWTH CGIPP GROWTH CVIC GROWTH CVIC GROWTH CVIC GROWTH GTR GROWTH GTR GROWTH GTR GROWTH TAS GROWTH TAS GROWTH TAS GROWTH WA GROWTH WA GROWTH WA GROWTH MVAL GROWTH MVAL GROWTH DOTHI GROWTH DOTHI DENSITY DENSITY 13+ DOTHI STEMST BRA BRQ BRS MOE

8 Schneeberger transformation BOT EBV u BOT = G BOT,SC G -1 SC,SC u SC

9 Radiata breeding was mostly OP in the 1 st generation, CP in the 2 nd and 3 rd generation now coming on line Generation Number of CP families Average CP family size Numbe r of CP parents Average number of mates per CP parent Number of OP families Average OP family size Numbe r of OP parents Number of genotypes Number of selections Selection ratio 0 30, , , ,

10 Lots of stem straightness scores, lots of branch scoring, majority of growth measurements in GTR. Generation 3 starting to come on line

11 Simulate the actual program Did not bother simulating non-test material (gen 0) OP to generate 1 st gen, CP to generate 2 nd gen Simulated a completed 3 rd gen (CP, same as 2 nd ) Number of genotypes, parents, mates per parent, family sizes were approximately the same as what has occurred in reality Approximately the same selection intensities Mirrored the true numbers of measurements by trait and by generation Used an average h 2 based on actual trial data Used an average error correlation matrix Used the current SC X SC matrix of correlations

12 Generation 2 Generation 1 Generation founders from different genetic groups (genetic group variance = 20% of the additive) Source 60 OP seed from each founder (2 founders in each group produce excess seed and act as link parents) Each OP family is tested in 3 trials (20 reps) 90 OP trials in total Number of genotypes measured for each trait same as reality 48 cols 1000 progeny from 50 parents 1160 progeny from 58 link parents 2160 progeny in total 45 rows TREEPLAN analysis and select top ranked 600 genotypes Each parent in 6 crosses (1800 families), 60 progeny per cross Each CP family tested in 3 trials (20 reps) 50 CP trials in total 48 cols Each CP trial tests 108 families each with 20 reps = rows

13 Scenario 1: run TREEPLAN with a diagonal G SC,SC matrix (equivalent to 27 single-trait runs) GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH DENSITY DENSITY DOTHI STEM CGIPP CGIPP CGIPP CVIC CVIC CVIC GTR GTR GTR TAS TAS TAS WA WA WA MVAL MVAL DOTHI DOTHI ST BRA BRQ BRS MOE GROWTH CGIPP GROWTH CGIPP GROWTH CGIPP GROWTH CVIC GROWTH CVIC GROWTH CVIC GROWTH GTR GROWTH GTR GROWTH GTR GROWTH TAS GROWTH TAS GROWTH TAS GROWTH WA GROWTH WA GROWTH WA GROWTH MVAL GROWTH MVAL GROWTH DOTHI GROWTH DOTHI DENSITY DENSITY 13+ DOTHI STEMST Zero correlation Zero correlation BRA BRQ BRS MOE

14 Scenario 2: run TREEPLAN with correlations among GROWTH SC unchanged but all other correlations = 0 GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH GROWTH DENSITY DENSITY DOTHI STEM CGIPP CGIPP CGIPP CVIC CVIC CVIC GTR GTR GTR TAS TAS TAS WA WA WA MVAL MVAL DOTHI DOTHI ST BRA BRQ BRS MOE GROWTH CGIPP GROWTH CGIPP GROWTH CGIPP GROWTH CVIC GROWTH CVIC GROWTH CVIC GROWTH GTR GROWTH GTR GROWTH GTR GROWTH TAS GROWTH TAS GROWTH TAS GROWTH WA GROWTH WA GROWTH WA GROWTH MVAL GROWTH MVAL GROWTH DOTHI GROWTH DOTHI DENSITY DENSITY 13+ DOTHI STEMST BRA BRQ BRS MOE

15 Mean NPV ($/m3) of the top 5% of the population

16

17

18 Conclusions For our situation (important traits measured on a small subset of trees) large multi-variate BLUPs are worth the effort We really should be increasing the number of measurements for these traits