Wednesday, October 26, 2016

NBCEC Brown Bagger: Implementation of single step methodologies at International Genetic Solutions

Dr. Mahdi Saatchi
International Genetic Solutions

IGS performs genetic evaluations for 12 breed associations from North America.
IGS has over 16 million animals in their database and is adding over 400,000 animals per yer.
IGS has 84,197 animals with genotypes. Simmental makes up about 40,000 of these genotypes.

Currently at IGS they blend the molecular breeding value (MBV, the genomic prediction) with the multi-breed international cattle evaluation. This is more like the blending that occurs with selection indexes.

Single-step genomic prediction allows information from genotyped animals to be spread to related animals in the data set.
Also, multiple-step genomic predictions were often trained on breeding values, and any errors in the estimation of the breeding values influenced the genomic prediction.

There are two approaches to single-step genetic evaluation. Single-step BLUP uses a breeding value model. Single-step Bayesian Regression uses a marker effects model.

In Single-step Bayesian Regression does not requiring inverting the relatedness matrix.
Also, in single-step Bayesian regression allows us to give different importance to DNA markers used (i.e. variable selection). In the dairy industry, they observe little benefit from Bayesian regression models. But, in the beef industry, we see genomic regions (i.e. QTLs) that have large effects in multiple breeds. Bayesian regression allows us to fit large effects to certain DNA markers used in the genetic prediction.

In the single-step Bayesian regression model, we infer (in other words predicting or imputing) the genotypes for all animals in the data set, even those animals that have not been genotyped.

Researchers at Iowa State, lead by Rohan Fernando, have created what they call a hybrid model which estimates marker effects for genotyped animals and breeding values for animals without genotypes.

With Simmental data, Bruce Golden see extra improvement in the precision when using a single-step Bayesian regression with different weights for DNA variants. This approach nearly cuts in half the uncertainty of EPDs compared with pedigree estimates.

This approach appears to be a major step forward in the precision of EPD estimation.


Post a Comment