arxiv: 2604.27831 · v1 · submitted 2026-04-30 · 📊 stat.AP

Recognition: unknown

Optimal allocation of trials to sub-regions in crop variety testing with multiple years and correlated genotype effects

Hans-Peter Piepho, Lenka Filov\'a, Maryna Prus, Waqas Ahmed Malik

Pith reviewed 2026-05-07 06:48 UTC · model grok-4.3

classification 📊 stat.AP

keywords variety testingoptimal allocationmixed modelskinshipsub-regionsbest linear unbiased predictiontrial designcrop breeding

0 comments

The pith

A method determines the optimal split of a fixed trial budget across sub-regions for crop variety testing when hundreds of genotypes are involved and kinship data are available.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Crop variety trials are run across sub-regions of a target environment population to predict how each genotype will perform locally. The authors treat genotype effects within each sub-region as random and build the variance-covariance matrix of these effects from pedigree or kinship information so that best linear unbiased prediction can borrow strength across zones. They combine closed-form results with numerical optimization to find the allocation of trials that best serves the prediction goal under a fixed total number of trials. The procedure remains computationally feasible when the number of genotypes reaches the hundreds, the scale common in commercial breeding programs. A reader would care because the same resources can then be used to produce sharper local recommendations instead of being spread evenly or chosen by rule of thumb.

Core claim

The optimal allocation of a fixed number of trials to sub-regions is obtained by solving a design problem that uses a kinship-derived variance-covariance matrix inside a mixed model for genotype-zone effects; the solution is reached through a combination of explicit analytic equations and numerical optimization and remains practical when the number of genotypes is in the hundreds.

What carries the argument

Mixed model with a variance-covariance matrix for genotype-zone effects constructed from kinship information, used to compute best linear unbiased predictions while optimizing trial allocation across zones.

If this is right

Prediction accuracy for genotype performance inside each sub-region improves for the same total number of trials.
The design remains tractable when the number of genotypes reaches several hundred.
Pedigree or kinship data can be fed directly into the allocation decision rather than used only after data collection.
Allocations automatically reflect the degree of correlation between zones instead of treating zones as independent.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Breeding programs could re-optimize allocations each year once updated kinship estimates become available.
The same framework might be applied to other multi-environment testing settings where spatial or genetic correlations are known.
Replacing pedigree kinship with genomic relationship matrices would be a direct next step that could sharpen the allocations further.

Load-bearing premise

A variance-covariance matrix built from kinship information accurately captures the pattern of genotype-zone interactions and yields reliable predictions for the purpose of choosing the design.

What would settle it

Run a simulation or field study in which the true genotype-zone correlations differ from those implied by the kinship matrix; if the prediction accuracy under the optimized allocation is no better than under equal allocation, the practical value of the method is refuted.

read the original abstract

Plant breeding and variety trials are usually conducted in multiple environments sampled from a defined target population of environments in order to characterize the performance of breeding lines or varieties. When the population is large and heterogeneous, it may be sub-divided into sub-regions or zones according to administrative and agro-ecological criteria. Analysis then focuses on prediction of performance in the individual sub-regions. Modelling the genotype effect in each sub-region as random, information can be borrowed across sub-regions using best linear unbiased prediction based on a suitable variance-covariance matrix for the genotype-zone effects. Here, we consider the important case where kinship of pedigree information is available for the genotypes under test. This information can be integrated into the variance-covariance matrix for genotype-zone effects. The objective we pursue here is to determine the optimal allocation of a fixed budget of trials to sub-regions. This design problem is solved using a combination of theory and explicit equations on one hand and numerical optimization on the other hand. Our proposed novel approach allows obtaining the optimal allocation when the number of genotypes is in the hundreds, a common setting in large plant breeding programs as well as in variety testing for economically important crops.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper provides a practical optimization method for allocating trials to sub-regions in variety testing by incorporating kinship into the genotype covariance matrix, though its effectiveness depends on that matrix being well-specified.

read the letter

The paper's key move is to treat genotype effects in different sub-regions as correlated random effects whose covariance comes from kinship information, then optimize the number of trials per region under a fixed total budget using a mix of closed-form expressions and numerical search. This lets them handle hundreds of genotypes without the computation blowing up. The title also mentions multiple years, so they probably extend the model to account for year effects and correlations over time as well. They do a good job laying out the model and showing how the prediction error variances depend on the allocation. The numerical optimization part seems to make it feasible for real breeding programs where you can't just run exhaustive search. Using the kinship to borrow information across zones is a natural step when you have pedigree data anyway. The main limitation is that everything rests on the kinship matrix giving a decent approximation to the true genotype-by-zone covariance. The paper probably derives the criterion correctly under that assumption, but without seeing checks against misspecified kinship or comparisons to simpler models, it's hard to know how much the recommended allocations would change if the correlations are off. The stress test note raises a fair point about sensitivity to the VCV; if the full paper doesn't include some robustness checks or real-data application that shows stable allocations, that would be a gap worth noting in review. Also, I wonder how they handle the multiple years part in the optimization – whether the allocation is per year or overall. This is aimed at statisticians working on variety trials and plant breeders who need to plan multi-location tests efficiently. Readers who already use BLUP and pedigree data in their analyses will get the most out of it. It could be a useful reference for designing experiments in heterogeneous environments. I think it deserves peer review. The core idea is clear and the practical angle is there, even if some details on robustness and the multi-year aspect need tightening. The claim about handling hundreds of genotypes is plausible given the mixed model setup.

Referee Report

2 major / 2 minor

Summary. The paper develops a method for optimally allocating a fixed budget of trials across sub-regions (zones) in multi-environment crop variety trials. Genotype effects within each zone are treated as random, with a variance-covariance matrix for genotype-by-zone interactions constructed from kinship/pedigree information; BLUPs are then used to borrow strength across zones. The design problem is solved by deriving analytic expressions for the prediction-error variance criterion and combining them with numerical optimization, with the explicit goal of scaling to hundreds of genotypes.

Significance. If the mixed-model predictions under the kinship-derived VCV are reliable, the approach could improve resource allocation in large-scale variety testing programs for heterogeneous target populations. The ability to handle realistic numbers of genotypes via a hybrid analytic-numerical strategy is a practical strength, and the integration of pedigree data into the design criterion is a clear advance over purely empirical or zone-independent allocations.

major comments (2)

[§3] §3 (Optimization criterion and algorithm): The central claim that optimal allocations can be obtained for hundreds of genotypes rests on the numerical optimizer converging to a global minimum of the PEV-based criterion under the assumed VCV. No convergence diagnostics, multiple-start results, or scaling experiments with genotype count are reported, leaving open whether the reported optima are stable or merely local for realistic program sizes.
[§4] §4 (Results and validation): No simulation study or sensitivity analysis examines how misspecification of the kinship-derived genotype-zone covariance matrix propagates into the optimized trial numbers per zone. Because the design criterion is a direct function of this VCV, even moderate errors in the estimated matrix can shift the numerical optimum substantially while still appearing optimal under the fitted model; this is load-bearing for the practical claim.

minor comments (2)

[Abstract] The abstract states that the method uses 'a combination of theory and explicit equations' but provides none; a single illustrative equation for the allocation criterion would improve readability without lengthening the abstract.
[§2] Notation for the genotype-zone covariance matrix (presumably denoted something like G ⊗ K or similar) should be introduced once in §2 and used consistently; occasional switches between 'kinship matrix' and 'VCV' are mildly confusing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and insightful comments on our manuscript. We appreciate the recognition of the practical strengths of our hybrid analytic-numerical approach for large-scale genotype sets. Below we provide point-by-point responses to the major comments. We have revised the manuscript to address the concerns raised regarding optimization reliability and sensitivity to covariance misspecification.

read point-by-point responses

Referee: §3 (Optimization criterion and algorithm): The central claim that optimal allocations can be obtained for hundreds of genotypes rests on the numerical optimizer converging to a global minimum of the PEV-based criterion under the assumed VCV. No convergence diagnostics, multiple-start results, or scaling experiments with genotype count are reported, leaving open whether the reported optima are stable or merely local for realistic program sizes.

Authors: We agree that additional evidence of optimizer reliability is necessary to support the claim of scalability to hundreds of genotypes. In the revised manuscript we will expand Section 3 to include: (i) results from 20 independent random initializations of the numerical optimizer for each reported allocation, (ii) convergence diagnostics (objective value trajectories and gradient norms) for representative problems, and (iii) scaling experiments that report wall-clock time, number of function evaluations, and stability of the resulting allocations as the number of genotypes increases from 50 to 500. These additions will demonstrate that the reported optima are robust rather than local artifacts. revision: yes
Referee: §4 (Results and validation): No simulation study or sensitivity analysis examines how misspecification of the kinship-derived genotype-zone covariance matrix propagates into the optimized trial numbers per zone. Because the design criterion is a direct function of this VCV, even moderate errors in the estimated matrix can shift the numerical optimum substantially while still appearing optimal under the fitted model; this is load-bearing for the practical claim.

Authors: The referee is correct that the dependence of the PEV criterion on the estimated VCV makes sensitivity to misspecification an important practical consideration. While a comprehensive Monte Carlo study assuming a different true VCV would be computationally demanding, we will add a targeted sensitivity analysis in the revised Section 4. Specifically, we will perturb the estimated kinship-derived VCV by multiplicative noise at relative levels of 5%, 10%, and 20% (reflecting plausible estimation error), re-optimize the allocations, and quantify the resulting changes in zone-level trial numbers and in the achieved PEV. The results will be summarized in a new table and accompanying text to illustrate the robustness of the optimal allocations. revision: yes

Circularity Check

0 steps flagged

No circularity: optimal allocation derived via external kinship matrix and numerical optimization of PEV criterion

full rationale

The derivation combines explicit analytic expressions for the mixed-model prediction-error variance matrix (under a kinship-derived VCV for genotype-zone effects) with numerical optimization of the resulting design criterion. The VCV is constructed from externally supplied pedigree/kinship data rather than estimated from the trial data whose allocation is being optimized; the objective function is therefore not fitted to the same observations used for evaluation. No step reduces the claimed optimal allocation to a fitted parameter, self-defined quantity, or load-bearing self-citation chain. The approach is a standard optimization problem under an assumed linear mixed model and remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard mixed-model assumptions in plant breeding statistics plus the feasibility of numerical optimization for the design criterion.

axioms (2)

domain assumption Genotype effects across sub-regions can be modeled as random variables whose variance-covariance matrix is informed by kinship or pedigree information.
Explicitly invoked in the abstract as the basis for borrowing information via BLUP.
domain assumption The target population of environments can be meaningfully partitioned into sub-regions for which separate predictions are desired.
Stated as the setting in which analysis focuses on individual sub-regions.

pith-pipeline@v0.9.0 · 5522 in / 1387 out tokens · 81369 ms · 2026-05-07T06:48:06.695010+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references

[1]

Albrecht, T., Wimmer, V., Auinger, H.-J., Erbe, M., Knaak, C., Ouzunova, M., Simianer, H., and Schön, C.-C. (2011). Genome-based prediction of testcross values in maize. Theoretical and Applied Genetics,123(2), 339–350. 23 Atkinson, A. C., Donev, A. N., and Tobias, R. D. (2007).Optimum Experimental Designs, with SAS. Oxford University Press, Oxford. Atlin...

2011
[2]

VanRaden, P. M. (2008). Efficient methods to compute genomic predictions.Journal of Dairy Science,91(11), 4414–4423. 26 Table 1:Optimal and highly efficient designs for standard and weightedA-criteria for prediction of genotype effects and their pairwise linear contrasts in models (1) and (2) for three kinship matrices withK= 31,K= 80, andK= 698genotypes ...

2008