Association Analysis > Genomic Selection: Error 260 using Tassel Tutorial Data (Tassel 5.2.93, MacOS)

[Jc Pineda]

I tried to run the Genomic Selection analysis using the Tassel Tutorial Data provided in this page. I am using MacOS.

According to the docs, the following files can be uploaded:

• kinship matrix (mdp_kinship.txt)
• phenotype (mdp_phenotype.csv)
• genotype (mdp_genotype.vcf)

It is not officially in the user manual, but I read that the genotype data has to be converted to numeric before running Genomic Selection (ref). So, I tried to convert the VCF to numerical by Data > Numerical Genotype.

After running Analysis > Association > Genomic Selection, I got a 206 Error Message. 

Did I miss a step? 

If anyone could guide me how to successfully run Genomic Selection using VCF files that would be great.

Thank you very much

[Peter Bradbury]

The help tab for Genomic Selection states " Predicts phenotypes using G-BLUP for genomic selection using a user-inputted kinship matrix and phenotype(s)", which is what you need, though mdp_phenotypes will not work for that (because the taxa are replicated). Instead use mdp_traits.txt and the kinship matrix. Select both in the left hand panel then start Genomic Selection. You get different outputs depending on whether cross-validation is checked, so try both ways.

Peter

[Peter Bradbury]

To use a VCF file just import it and run Analysis/Relatedness/Kinship to create a kinship matrix. The reference you cited was related to what the User needed to do to import a certain type of data. You do not need to convert a VCF to numeric data.

[Jc Pineda]

Hi Peter,

Thank you for your response.

I confirm Genomic Selection works using mdp_traits and mdp_kinship.

Pardon for my lack of understanding, but I'd like to clarify a few points point about Tassel's Genomic Selection function. 

• The docs indeed state that Genomic Selection predicts phenotype using kinship matrix and phenotypes. However, the output was only a summary table of the prediction accuracy (mean and standard deviation). Is there another step required to predict phenotypes or GEBVs in Tassel? 
• Doesn't phenotype prediction require genotype / SNP data as input? In our workflow, we usually estimate the GEBV of a breeding population, wherein individuals only have genotypic information.

[Peter Bradbury]

Accuracy_mdp_traits contains the results for each fold x iteration and can be exported and plotted using R or other software. Those results show up on the right side of the screen. If you uncheck "Perform cross-validation", then the predictions with PEVs (prediction error variance) for each taxon in the kinship matrix. To get predictions for taxa that do not have phenotypes, include those in the kinship matrix along with the training set that does have phenotypes.

Peter