Abstract
Epistasis is the interaction between loci, which has been recognized as having a critical role in dissecting genetic control of physiological pathways and as an important source of genetic variance in plant populations.
The objective was to explore epistatic genetic effects (EPI) in simulated double haploid (DH) populations. The impacts of EPI on genomic prediction (GP) of total (additive + epistatic) genetic effects were studied in a half diallel using different designs where different parts as reference to predict the remaining crosses (RDs).
We simulated data from 2,280 DHs originated from 190 F1 families generated by half diallel cross between 20 homozygous diploid parents. Three different RDs were investigated by varying the structure of reference population: 1) 10 families with no overlap of parents among different families (RD1), 2) 40 families with one parent overlap in two neighboring families (RD2), and 3) 120 randomly selected families accounted for ~ 60% of whole population (RD3).
In total, five different statistic models/methods were applied for GP. The models were two genomic best linear unbiased prediction (GBLUP) models with EPI, one GBLUP model without EPI, a single nucleotide polymorphism BLUP model without EPI, and a reproducing kernel Hilbert space regression model capturing EPI. The predictive ability (PA) was measured as the correlation between true and estimated total genetic values.
Results showed that RD2 generally provided higher PA than RD1 and RD3. The influence of the EPI was not observed in PA since there was no significant difference between models with or without EPI.
The objective was to explore epistatic genetic effects (EPI) in simulated double haploid (DH) populations. The impacts of EPI on genomic prediction (GP) of total (additive + epistatic) genetic effects were studied in a half diallel using different designs where different parts as reference to predict the remaining crosses (RDs).
We simulated data from 2,280 DHs originated from 190 F1 families generated by half diallel cross between 20 homozygous diploid parents. Three different RDs were investigated by varying the structure of reference population: 1) 10 families with no overlap of parents among different families (RD1), 2) 40 families with one parent overlap in two neighboring families (RD2), and 3) 120 randomly selected families accounted for ~ 60% of whole population (RD3).
In total, five different statistic models/methods were applied for GP. The models were two genomic best linear unbiased prediction (GBLUP) models with EPI, one GBLUP model without EPI, a single nucleotide polymorphism BLUP model without EPI, and a reproducing kernel Hilbert space regression model capturing EPI. The predictive ability (PA) was measured as the correlation between true and estimated total genetic values.
Results showed that RD2 generally provided higher PA than RD1 and RD3. The influence of the EPI was not observed in PA since there was no significant difference between models with or without EPI.
| Originalsprog | Engelsk |
|---|---|
| Publikationsdato | 2019 |
| Status | Udgivet - 2019 |
| Begivenhed | The Plant and Animal Genome XXVII Conference (PAG) 2019 - San Diego, USA Varighed: 12 jan. 2019 → 16 jan. 2019 Konferencens nummer: XXVII |
Konference
| Konference | The Plant and Animal Genome XXVII Conference (PAG) 2019 |
|---|---|
| Nummer | XXVII |
| Land/Område | USA |
| By | San Diego |
| Periode | 12/01/2019 → 16/01/2019 |