We don't ship vanity metrics. Below are anonymized but quantitatively faithful summaries of breeding teams running SEED at scale — prediction accuracies, p-values, and time-to-decision.
Maize · Regional breeding program, Sub-Saharan Africa
Drought-tolerance GWAS in a 1,240-line maize panel
Significant SNPs (Bonferroni)
17
Genomic inflation λ
1.04
Top hit −log₁₀(p)
9.8
Candidate genes
23 annotated
The team ran MLM-GWAS with 3 PCs as covariates on field-measured anthesis-silking interval under managed drought. SEED surfaced 17 Bonferroni-significant SNPs across chromosomes 1, 3, and 8 — the chromosome 3 cluster overlapped previously published QTL for ASI. Gene annotation tied the lead SNP to a dehydration-responsive transcription factor.
Wheat · Public breeding program, North America
Genomic selection cross-validation for grain protein
Training set size
2,180 lines
GBLUP CV accuracy (r)
0.62
Heritability (h²)
0.71
Cycles saved per release
~2
The program replaced phenotyping every line with GS-based ranking. Five-fold cross-validated accuracy on grain protein hit r = 0.62, comfortably above the threshold to deploy. By selecting on GEBVs instead of measured protein, the team trimmed roughly two cycles off the time-to-release without sacrificing genetic gain.
Multi-environment trial decomposition across 9 locations
Genotypes × Environments
84 × 9
AMMI PC1 variance
41%
Stable winners identified
6
Mega-environments
3
SEED's MET module decomposed yield into main effects and GxE, with AMMI PC1 capturing 41% of the interaction. Finlay–Wilkinson slopes identified six broadly stable genotypes (b ≈ 1, low Wricke ecovalence). The team re-zoned the variety portfolio across three mega-environments — a decision that previously took an entire off-season meeting.
Bring your data, get real numbers
Every metric above came from the same modules you'd run on your own dataset.