HGATGS: Hypergraph Attention Network for Crop Genomic Selection

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this paper, He et al. developed a novel crop genome selection model based on hypergraph attention networks for genomic prediction (HGATGS). The outline of the models appear to be clear. The results show that HGATGS outperformed traditional statistical methods and machine learning models across various datasets, however, I have identified significant concerns regarding the results of the study.

Firstly, the authors calculated cosine distances between all samples to measure similarity. Cosine distance evaluates the angular difference between two vectors, which raises concerns about how collinearity between features was addressed. The paper does not provide explanation or methodology for handling this issue.

Secondly, discrepancies in the reported results are evident. Table 2 presents the prediction correlation for each trait, while Table 3 provides the average prediction correlation across all traits. The average correlations for Wheat 599, Wheat 2000, and Wheat 487 appear consistent. However, the average correlations for G2F 2017 are significantly higher than any individual trait, while those for Rice 299 are much lower. These inconsistencies undermine the reliability of the results.

Additionally, the correlations reported in Table 2 do not align with Figure 3C, and those in Table 3 do not match Figure 4. This inconsistency between tabular data and graphical representations further detracts from the paper's credibility.

Finally, the script and source data are not available in the Data Availability statement, making it impossible to verify the findings or reproduce the analyses. For instance, without access to the dataset and script, it remains unclear how the training and test sets were split during K-fold cross-validation, particularly in the context of multi-environment trials.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript "HGATGS: A Hypergraph Attention Network-Based Method for Crop Genomic Selection” proposes a crop genome selection model based on hypergraph attention networks for genomic prediction, named HGATGS. Some conventional statistical and machine learning models were used on five datasets from wheat, maize, and rice to evaluate the predictive capabilities of the proposed HGATGS model. Interestingly, the results across the datasets show that the model consistently delivered exceptional performance in predicting multiple traits.

The experimental strategy was well executed, and the article is well written. However, some points require improvement in the presentation of the results:

1) Please review the following text and rewrite it according to the results. Lines 357-359: “Specifically, in the Wheat 487 dataset, HGATGS showed outstanding performance in predicting traits such as TKW and GW, achieving correlation coefficients of 0.74 and 0.75, respectively.”.

2) The statements in lines 442-452 do not seem to make much sense considering the results shown in Table 4, where the GCN model does not seem to have performed that badly compared to the other models.

3) The assertion in lines 457-458 appear to be incomplete, considering that the same result occurs in all three datasets.

4) The text in lines 460-461 "achieved the best performance across the three datasets, with correlation coefficients of 0.53 and 0.67, respectively” should be "achieved the best performance across the three datasets, with correlation coefficients of 0.53, 0.67, and 0.9, respectively”.

Comments on the Quality of English Language

The text requires minor corrections and improvements, and I recommend conducting a thorough review.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed my previous concerns and the manuscript has greatly improved. Well done!