September 7th, 2020
The scope of the simulations to come: accuracy
The scope of the simulations to come: intensity
The scope of the simulations to come: cycle time (skipping stages in PD)
The scope of the simulations to come: cycle time (early recycling)
1. Introduction to the problem
Crop by region
CIAT-Beans EA
Problem specification
Maximum accuracy is not being exploted to increase rates of genetic gain.
Breeding strategy component tackled
Crossing, Evaluation, Selection
Breeders’ equation terms tackled
r
\(\Delta_g = (i * \sigma_g * r)/L\)
Hypothesis
Using GBLUP in the selection process to increase accuracy may increase the rate genetic gain.
2. Materials and methods
Treatments
| Treatment | Description |
|---|---|
| PYTF7_VEF_OF_BLUP | Current scheme, selecting parents at PYT_F7 (20), VEF_F8 (5), OF_F9 (10) using a base index. |
| PYTF5_VEF_OF_FAM_BLUP | Scheme selecting parents at PYT_F5 (20), VEF_F6 (5), OF_F7 (10) + top 20% families are selected at F2 using a base index (not parents) using BLUPs. |
| PYTF5_VEF_OF_FAM_PBLUP | Scheme selecting parents at PYT_F5 (20), VEF_F6 (5), OF_F7 (10) + top 20% families are selected at F2 using a base index (not parents) using PBLUPs. |
| PYTF5_VEF_OF_FAM_GBLUP | Scheme selecting parents at PYT_F5 (20), VEF_F6 (5), OF_F7 (10) + top 20% families are selected at F2 using a base index (not parents) using GBLUPs. |
Simulation procedure
A 20 year burn-in period was modeled using the current breeding scheme. The burn-in was followed by a 30 year evaluation period to measure rates of genetic gain for all treatments. Genetic gain was measured by assessing changes in genetic merit in F9 lines. Genotype-by-year interaction variance was assumed to be equivalent to genetic variance (based on average correlation between locations being equal to 0.5). 20 replications done. We simulated 5 complex and 3 simple traits to be behind the genetic merit.
The targeted change
The different models to estimate the surrogates of genetic value are applied at the recycling stages (PYT & VEF).
PYT: 3reps; 1env; h2=0.54; r=0.73
VEF: 3reps; 2env; h2=0.54; r=0.73
3. Results in CIAT-Beans (genetic gain)
- We knew that starting PYT at F5 instead of F7 provides 1.26 (95% CI: 1.23,1.29) times more gain.
- Changing the estimate method at PYT (3reps;h2=0.54; r=0.73) provides 1.02 (95% CI: 1.01,1.04) times more gain.
3. Results in terms of accuracy
PYT across years
The genetic model (r, R and \(\sigma_g\))
At low heritability the gain in accuracy is important. At intermediate and high heritabilities the gain in accuracy is marginal.
The genetic model (in % increase/decrease)
At low heritability the gain in response to selection is important. At intermediate and high heritabilities the gain in response to selection is marginal.
There’s an increase in accuacy (i.e. at PYT) in all years and traits, but not enough to change selection decisions drastically and lead to greater gains.
4. Conclusion
Using GBLUP to estimate the surrogate of genetic value for selection can increase the accuracy dramatically at low h2 EVEN for small family sizes.
Using GBLUP to estimate the surrogate of genetic value for selection has no effect at intermediate and high heritability trials. In the CIAT-Beans program using GBLUP at recycling stages:
PYT: 1 location 3 reps (h2=0.54; r=0.73 at each location)
VEF: 2 locations 3 reps (h2=0.54; r=0.73 at each location)
Increases the accuracy but doesn’t seem to change selection decision drastically to increase the rate of gain.
We recommend to use GBLUP to keep accuracy high even in the worst scenarios and to make it the first step to start using GS in the near future.