Modern breeding programs for species of economic and agricultural significance require estimates of breeding values for several traits per individual. Today's method of choice for estimation of breeding values for continuous traits is Best Linear Unbiased Prediction (BLUP) with an Animal Model as the operational model. This leads to the need to solve large systems of linear equations with at least as many unknowns as the number of individuals multiplied by the number of traits. For large populations and many traits, setting up and solving the equation systems exceed the capacity of a single workstation at least within a reasonable time frame. In species like cattle and pigs, data are recorded continuously and selection decisions, based on all available data, have to be made throughout the year, so updated (weekly or monthly) estimates are required. One way to overcome this computational challenge is to utilize parallel computation. The CEBUS project will develop a parallel solver for multivariate mixed models. The partners will adapt and use this software on a specific large and commercially significant example. This example will be used to demonstrate the benefits of an HPCN approach to animal breeding. Another result will be a new module in the DMU-software package, developed at DIAS. The DMU-packages are currently used in many countries for analysing animal breeding data. The resulting software package and its use on high-performance parallel systems will have significant benefit to the European livestock industry.
Project URL: http://industry.ebi.ac.uk/BioTitan/activities/cebus/cebus.html