Skip to content

Latest commit

 

History

History
93 lines (52 loc) · 2.83 KB

Selective Phenotyping.md

File metadata and controls

93 lines (52 loc) · 2.83 KB

E-GP: selective phenotyping

author: Germano Costa-Neto

date: May 2021

As a toy-example, we made available a set of 100 genotypes over 4 environments.

Functions and data

Functions for Enviromic Assembly came from the EnvRtype-package. The Supplementary data can be downloaded here

# packages
require(STPGA)
require(superheat)

# source codes
source('https://raw.githubusercontent.com/gcostaneto/EGP/main/supplementary_codes_2.R')

# supplementary data
load('Genomic_Enviromic.RData')  # genomic relatedness for A effects (100 x 100)

Definition of the Effective Number of Observations

First create the genotype-environment kernel by the kronecker product between K_ET and K_G;

superheat(K_E) # environmic kernel builted for some 4 environments
superheat(K_G) # genomic kernel with 100 genotypes

K_GE = kronecker(K_E,K_G,make.dimnames = T)
colnames(K_GE) # names of genotypes x environment combinations

Then idenditify the effective number of genotype-environment combinations. This algoritm is inspired in Misztal (2016), in which the effect number of genotypes using only K_G. By default, fraction is = 98% (0.98).

SVD = N_GE(Kernel = K_GE,plot = F,fraction = .98,svd.print = T)
ntrain = SVD$Ne        # N_GE
svdGEI = K_GE %*% SVD$svd.v # SVD-based GEI matrix for optimize the genetic algorithm
rownames(svdGEI) = rownames(K_GE)

Genetic Algorithm for Selecting MET Subsets

Parameters:

100 iteraions (niterations);

80% of mutation for each generated solution (mutprob = 0.80)

5 solutions selected as elite parents for next generations (nelite = 5);

PEVmean criteria (errorstat = 'PEVMEAN')

TS <- GenAlgForSubsetSelectionNoTest(P = as.matrix(svdGEI), ntoselect = ntrain,
                                     nelite = 5,mutprob = .8, niterations = 100, plotiters = F, 
                                     lambda = 1e-5, errorstat = "PEVMEAN", mc.cores = 4)[[1]]


TS # training set (super-optmized) consisted by some genotypes at some environments
superheat(TS)

saveRDS(object = TS, file = paste0(ntrain,"_optmized_SET"))

Final Considerations

  • in the file TS *, it is presented genotype_by_environement combinations to compose a super-optimized training set for genomic prediction.
  • we recommend to check for your dataset which optimization parameters lead to better selective schemes;
  • we also strong recommend the use of different enviromic assembly according to the environmental complexity of you will be dealing with.
  • If you found some concerns or have any suggestions, please contact us.