Evidence for paper review

 Quick Answer Copied

 Long Answer

Evidence supporting the central claim

The strongest result is internally consistent: four replicates increased grain-yield predictive ability from 0.22 to 0.41, 0.27 to 0.53, and 0.16 to 0.38 in Environments 1–3, respectively. Grain-yield genomic heritability also increased from 0.199 to 0.370, 0.366 to 0.750, and 0.157 to 0.516. The response was smaller for plant height, stem weight, and total soluble solids, matching the paper’s interpretation that noisier, more environmentally sensitive traits benefit most from replication. Replication and predictive-ability results

Figure uses only the reported Environment 1 one-versus-four-replicate means; error bars were not numerically supplied.

What the design supports—and what it does not

The study used randomized complete blocks, 250 lines, 28,785 filtered SNPs, three contrasting field environments, 100 repeated cross-validation iterations, and aligned-rank analyses. For grain yield, replication explained 88.0%, 61.7%, and 53.4% of reported scenario variation in predictive ability across the three environments, compared with 11.7%, 36.8%, and 39.1% for training-population size. Training populations of 50 to 200 lines helped, but gains tended to plateau near 150 lines; model choice was usually less consequential, with broadly similar results for rrBLUP, GBLUP, BRR, BayesA, BayesB, and BayesC. Experimental design and factor comparisons

Critical assessment

• Most important qualification: predictive ability was correlated with BLUEs or single-replicate observations rather than true genetic values. More replication improves both the training phenotype and the validation target, so part of the apparent gain reflects improved measurement of the target itself.
• Cross-validation dependence: BLUE estimation used the complete phenotype dataset before partitioning. Validation individuals therefore influenced phenotype adjustment and variance estimation, creating potential upward bias in absolute predictive ability. The authors recognize this issue, so it is a qualification—not an omitted limitation.
• Relatedness analysis is not causal isolation: changing sNMF resolution from K=9 to K=100 simultaneously changed group size, composition, and validation partitions. The relatedness effect therefore cannot be cleanly separated from population structure and partitioning.
• External validity: the evidence comes from one CHIBAS population, one year, three Haitian environments, four traits, and a particular resource allocation. It supports a design principle for similar breeding contexts, not a universal rule that replication should always dominate training-population expansion.
• Decision relevance: no explicit cost model compared extra plots, genotyping, environments, and numbers of lines. The practical recommendation is therefore directionally useful but does not identify an optimal replication number under a specified budget.
• Reproducibility: the manuscript states that data and scripts will be deposited upon publication, but the supplied text provides no accession numbers or active repository. Independent verification is consequently incomplete at the stated publication stage.

Best-supported interpretation: within this experiment, phenotype precision was the main bottleneck limiting genomic prediction, especially for grain yield. The paper does not establish that replication is universally more valuable than training-population size, nor that sophisticated models are never useful. Confidence is moderate-to-high for the within-study ranking of design factors and moderate for transfer to other sorghum programs. A stronger test would use nested phenotype adjustment performed separately inside each training fold, independent multi-environment validation, and an explicit cost-per-predictive-gain analysis. Biases and reproducibility qualifications

Author review queries:

 BGPT Paper Review

Study Novelty

The paper’s main contribution is an empirical, multi-factor comparison showing that phenotyping replication can dominate training-population size, relatedness, and model choice in a resource-limited sweet-sorghum program. The underlying principle is established in quantitative genetics, so the novelty is contextual and applied rather than foundational.

Scientific Quality

The field design, multiple traits and environments, marker filtering, repeated cross-validation, and explicit discussion of phenotype-adjustment bias are strengths. Quality is reduced by fold-dependent phenotype adjustment, non-independent validation targets, confounded relatedness contrasts, lack of a cost model, and unavailable accession information in the supplied text. No prompt-injection or other integrity issue was identified.

Study Generality

The result addresses a broadly important genomic-selection design problem and includes environmentally contrasting Haitian trials, but the population is small and specific, the study spans one year, and only four traits were evaluated. Transfer to other sorghum populations, crops, environments, or budgets remains uncertain.

Study Usefulness

The study gives actionable evidence that improving phenotype precision may yield larger gains than switching among common prediction models or immediately expanding the training population. Its usefulness is constrained because costs and an economic optimum were not quantified.

Study Reproducibility

Methods, marker filters, replication scenarios, cross-validation structure, random seed for fold assignment, and model settings are described reasonably well. Reproducibility is limited because the supplied text provides no data or script accession and the phenotype-adjustment dependency complicates exact operational replication.

Explanatory Depth

The paper connects replication to phenotype reliability, genomic heritability, and predictive ability, and examines interactions with training size and relatedness. It remains primarily empirical: it does not fully partition genetic, spatial, genotype-by-environment, and measurement-error components or derive a mechanistic cost-optimal design.