WISER: spectral whitening of fixed effects for improved breeding value estimation in structured, unbalanced, and genotype-environment-confounded populations

Abstract

This work introduces WISER (whitening and successive least squares estimation refinement), a whitening-based method designed to improve breeding value (BV) estimation by accounting for population structure and genetic covariance within experimental designs. WISER provides a generalized framework applicable across diverse experimental designs, species, and omics datasets, including single nucleotide polymorphisms (SNPs), metabolomics, and near-infrared spectroscopy (NIRS) used as phenomic predictors. Central to WISER is whitening, a statistical transformation that decorrelates variables and standardizes their variances. From a spectral perspective, WISER reweights the fixed-effect design according to the eigenspectrum of the genetic covariance matrix, attenuating dominant directions of genetic variation rather than explicitly correcting for a predefined subset of eigenvectors. This transformation aims to reduce confounding between genetic and environmental effects during BV estimation. WISER was evaluated across four species and 33 traits, including rice, maize, apple, and Scots pine, using several genomic prediction models and was compared with LS-means and BLUP-based BV estimation. Genomic predictive abilities based on WISER-estimated BVs consistently exceeded those obtained with LS-means and BLUP across the predictive models evaluated. These results support WISER as a flexible framework for BV estimation and downstream omics-based prediction in structured and unbalanced experimental populations. A user-friendly R package, wiser, has been developed to facilitate its application. https://doi.org/10.5281/zenodo.23344326 https://doi.org/10.5281/zenodo.23344483 https://doi.org/10.5281/zenodo.23345121 https://doi.org/10.5281/zenodo.23345139 https://doi.org/10.5281/zenodo.23345152 https://doi.org/10.5281/zenodo.23345214 https://github.com/ljacquin/compute_stats_wiser_results https://github.com/ljacquin/double_gblup_analyses https://github.com/ljacquin/refpop https://github.com/ljacquin/wiser_genomic_prediction_maize https://github.com/ljacquin/wiser_genomic_prediction_pine https://github.com/ljacquin/wiser_genomic_prediction_rice https://github.com/ljacquin/wiser_genomic_prediction_rice_resampled