Title: Statistical methods for genomic breeding value estimation, classification and performance prediction

Project manager: Prof. Dr. H.-P. Piepho (UHO-BI)

Projekt partners: KWS, TUM-PZ, UGÖ-TZ, UHO-BI

Objective:
Development and test of efficient statistical analysis and validation methods for genomic selection, breeding value estimation and performance prediction.

Summary:
In order to exploit new options provided by high-throughput techniques in plant and animal breeding, there is a need to develop new and efficient statistical methods. Modern marker techniques (SNPs), which allow a dense coverage of the genome, hold particular promise for genomic selection and breeding value estimation. This approach differs from previous approaches to QTL- and association mapping in that it abandons the objective to map individual genes and instead focuses on an efficient estimation of breeding values using all available marker information. The Synbreed project A1 develops parametric and non-parametric statistical procedures for genomic selection and breeding value estimation and adapts these to the specific needs for chicken, cattle and maize. A1 gives particular emphasis to semi- and non-parametric and machine learning methods. Compared to parametric approaches such as BLUP and ridge regression as well as Bayesian procedures, these have the advantage that non-additive gene effects are conveniently accounted for, without a need to make strong quantitative-genetic model assumptions, which in some cases are difficult to validate or verify. Based on mixed model procedures, extensions are proposed that allow incorporating non-genetic sources of variation, thus improving the predictive accuracy of genomic breeding value estimation. In order to assess the performance of alternative procedures, simulation and cross-validation procedures will be developed and made available to the project partners.

Sub-projects:
A1.1: Genomic breeding value estimation and classification
A1.2: Extended modelling for genomic breeding value estimation
A1.3: Cross-validation and calibration