Optimization in Breeding Programs Using Evolutionary Algorithms
Breeding programs aim at improving the genetic properties of livestock and crop populations with regard to productivity, fitness and adaptation. With the rise of genomics and new breeding technologies breeding programs have undergone significant transformations over the past few decades that allow for more complex and advanced designs. A multitude of variable and highly interdependent input parameters and many contrasting breeding goals significantly increase the challenge to derive an optimal breeding strategy. As result, the optimization of modern breeding programs involves many interdependent decisions on how to allocate resources, to obtain both short-term genetic and long-term sustainability. With increasing computational power, stochastic simulations have emerged as a valuable enhancement for analyzing breeding programs to predict multiple breeding decisions. For this purpose, a variety of software to perform stochastic simulations, such as MoBPS and AlphaSim were developed. However, analysis/optimization is usually limited to a couple of selected scenario and does not allow for optimization in the sense of the best use of resources, achieve optimal outcomes for a given budget and provide effective trade-offs between competing objectives. Since there is a significant amount of randomness in both, simulations and real-world breeding programs, there is the need for further optimization that allows to reduce the stochasticity in data patterns and improve the accuracy of outcome predictions.
Challenge
Stochastic simulation is usually limited to the comparison of a few selected scenarios. There is a need for a dynamic optimization for resource allocation in breeding programs using evolutionary algorithms. Parameters to optimize:
• Number of test daughters
• Number of test bulls
• Number of selected proven sires
• Search space / constraints
Our Solution
The general concept of stochastic simulations, kernel regression and evolutionary algorithms are already well established. However, evolutionary algorithm are very specific to an application so the concept is vague and allows/requires adaptation to the problem setting. So the design of the evolutionary algorithm is the main innovative step with many of steps in the evolutionary pipeline. The evolutionary algorithm used in addition to stochastic simulation with established models, such as MOBPS, with kernel regression, includes the following steps:
(1) Concept of combining meiosis (recombination/mutation) in regard to parametrizations of a breeding program (while fulfilling constrains of the optimization problem)
(2) Kernel regression as a tool to select parents parametrization
(3) Kernel regression as a tool to predict performance of parameter settings that where not simulated themselves
(4) Derive final optima based on all prior iterations
Advantages
Every breeding action in a breeding scheme is simulated. Results of a simulation will not be an expected gain but the realization of a stochastic process.
- Low number of assumptions
- Flexibility
- High level of detail
- High computational demands and work
Applications
This new method to optimize plant and animal breeding programs is primarily interesting for
- plant breeding companies. BASF, as a patent co-owner, is already using the method.
- animal breeding companies
Development Status
Ready to use
Patent Status
IP rights ( EP 24188636.5 A1) have been filed in the name of the University Göttingen and BASF. A licensing partner for application in animal field is sought.
References
https://doi.org/10.1093/g3journal/jkad217
https://doi.org/10.1534/g3.120.401193
https://doi.org/10.1093/g3journal/jkab023
Contact
Christiane Bolli
Patent Manager lifescience
E-Mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
Tel.: +49 551 30724 156
Reference: BioM-2560-SUG
Tags: Algorithm, Animal Breeding, Plant Breeding
