The Problem
Breeding describes the process of selecting animals or plants with favorable characteristics that can be inherited to improve future generations and obtain higher yields, improved quality, disease resistance, and sustainability. To do this, characteristics (“phenotypes”) and DNA information (“genotypes”) of individuals are collected and subsequently used to estimate the expected performance of an offspring (“breeding value”). As modern breeding programs can contain millions of individuals and tens of thousands of genetic markers, this results in high computational demands. Researchers, such as those at Wageningen University & Research (WUR), work on developing algorithms and applications for dealing with these challenges.
The Quantum Solution
To address these challenges, this research developed and explored together with WUR randomized algorithms for singular-value decompositions. By reducing the dimensionality of large genetic datasets, these methods speed up computational processes while preserving accuracy. A key component explored is the use of quantum-inspired sampling techniques to focus on retrieving the top 5% of individuals with the highest genetic merit, avoiding the inefficiency of estimating the genetic merit of the remaining 95% individuals. Tools like the truncated Singular Value Decomposition (SVD) and Halko’s randomized algorithms demonstrated significant potential in reducing computation times while maintaining acceptable accuracy.
The blue curve denotes the actual breeding values of specimens with the breeding values on the y-axis and on the ax-axis its position in ascending order. The red lines denote the sampled highest 5% breeding values using the quantum inspired algorithm.
The Benefits
The approach could offer major potential benefits for genomic selection. Dimensionality reduction techniques enable more efficient genomic predictions, optimizing resource use and reducing the time needed to identify individuals with high genetic merit. Early experiments validated that tools like Halko’s randomized SVD can accelerate computations while maintaining accuracy, making these methods particularly valuable for large-scale datasets, even without the use of quantum(-inspired) methods. By focusing computational efforts on high-value outcomes, quantum and quantum-inspired algorithms have the potential to enhance breeding programs, supporting greater food security, enhanced biodiversity conservation, and a more sustainable agricultural future.
This work is supported by the Dutch National Growth Fund (NGF), as part of the Quantum Delta NL programme
