Exploiting Low-Rank Structure in Genomic Relationship Matrices for Computationally Efficient Optimum Contribution Selection
Ahlinder, J.; Waldmann, P.
Show abstract
Motivation: Optimum contribution selection (OCS) maximizes genetic gain under a constraint on relatedness. Genomic relationship matrices are dense, so the quadratic term costs O(n^3) operations, and existing OCS software assumes pedigree sparsity. Results: We replace the relationship matrix by its leading eigenpairs, computed by randomized SVD, and solve OCS in factorized form. On Norway spruce population (n=5,525), ranks 10--30 kept the exact objective within 1% of the optimum at 110--330-fold speedups; on the QTL-MAS 2010 (n=3,226), ranks up to 100 were needed (24-211 fold). The required rank rises as the diversity constraint loosens. Gain alone is a misleading accuracy check, since coancestry exceeded the optimum by up to 28%, so we choose the rank on the exact objective, which can be checked without a dense solve. The speedup comes from computing only the leading eigenpairs. We extend the factorization to robust OCS (18-fold). Availability and implementation: Julia implementation, and examples are available at https://github.com/jonhar97/fastocs under the MIT license (archived at Zenodo, DOI 10.5281/zenodo.23214804).
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- miqoGraph : Fitting admixture graphs using mixed-integer quadratic optimization 93%
- Large-scale Inference of Population Structure in Presence of Missingness using PCA 93%
- KIMGENS: A novel method to estimate kinship in organisms with mixed haploid diploid genetic systems robust to population structure 93%
Similar papers in this journal
- Genomic prediction of individual inbreeding levels for the management of genetic diversity in populations with small effective size 93%
- Complex genetic admixture histories reconstructed with Approximate Bayesian Computations 92%
- Estimating the time since admixture from phased and unphased molecular data 92%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.