Hello,
I am using SAS version 9.4
I am analysing freezing tolerance in plants as measured by the percentage of electrolyte leakage (EL) after freezing treatment.
All the genotypes tested are siblings.
My experimental design is a split plot design with 5 replications.
My main factor is freezing treatment or temperature (3 levels) and my subfactor is plant genotype (100 levels).
In each experiment (or replication) I measured EL on 3 stem sections per genotype and per temperature.
I consider that stem section is a factor nested within genotype but I think that I still need to specify the residuals 'Block*FactorA*FactorB' in the model since I have multiple data point for each genotype*temperature.
My ultimate goal here is to compute BLUP's for each genotype and temperature across 5 replications, to be used in genetic mapping.
I consider temperature as a fixed effect, while I consider both replication and genotype as random effects in a mixed model.
Indeed, the 100 genotypes represent a sub-sample of the entire bi-parental population.
My response variable (EL) is a percentage, so I chose to indicate a beta distribution with the linked scale Logit using proc glimmix and laplace approach.
Please see my script below:
data first;
input Rep$ Temperature$ Genotype$ Stem_Section$ EL;
datalines;
[...]
proc glimmix data = first method=laplace;
EL = EL/100;
class Temperature Genotype Rep Stem_Section;
model EL= Temperature/ dist = beta link=logit ;
random Rep Rep*Temperature Genotype Genotype*Temperature Rep*Temperature*Genotype / solution;
output out=second pred(ilink blup)=pblup ;
proc print data=second;
title 'Mixed model analysis with Random Genotype effect computing BLUPs'
Run;
In the random statement, 'Rep Rep*Temperature Genotype Genotype*Temperature' specify the G matrice
while 'Rep*Temperature*Genotype' specifies the residuals associated with the presence of multiple measurements and then the R matrice.
With 100 genotypes, SAS never ends processing the data.
I have no problems when I run the model with only 5 genotypes and I obtain Subject=1 in the Dimensions table, which makes sense to me since I have specified both the G and R matrices.
I also obtain blup estimates for Genotype*Temperature*Rep.
I would then pursue by calculating the mean of Blup's per genotype and temperature across 5 replications to obtain Blups for each genotype and temperature to be used in mapping.
My questions are:
1) Is my approach for estimating BLUP's correct and did I stated my random effects properly without overspecifying my split plot design?
2) I already increased my SAS memory, how can I process all 100 genotypes at once to compute BLUP's using SAS?
Thank you so much for any help you could provide.
Sincerely,
Daisy