All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
All of the reviewers' comments have been addressed by the authors. The revised manuscript can be accepted for publication.
[# PeerJ Staff Note - this decision was reviewed and approved by Shawn Gomez, a PeerJ Section Editor covering this Section #]
The writing is clear and unambiguous. The manuscript meets the basic reporting requirements.
The authors have made major revisions to the manuscript, including more details about their molecular representation method and performing more experiments to compare their model with other LLM-based and graph neural network models. The experiment design is good, and the new results support their method.
New visualizations are helpful for the validation of the proposed model. I have no further comment.
No further comments. Everything looks good.
- The revised manuscript now has sufficient information and is well-structured. The addition of a visualized data preprocessing pipeline and violin plots enhances clarity and readability.
- Background and references are comprehensive and appropriately contextualized within DILI prediction research.
- The GAT+GCN architecture and preprocessing workflow are clearly described and well-illustrated.
- Visual explanations effectively highlight key molecular substructures, improving interpretability.
- Comparative results and distributional analyses demonstrate robustness and reproducibility across multiple runs.
- Code, data, and experimental settings are sufficiently documented for independent verification.
- The conclusions are well-supported by the presented results and align with the study’s objectives.
Please made a revision for the manuscript based on the comments from the reviewers. The authors should consider comparing the proposed model with other machine learning and deep learning models listed in the comments.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
The authors present an interesting application of hybrid graph neural networks, specifically combining Graph Attention Networks (GAT) and Graph Convolutional Networks (GCN), for DILI compound prediction. The integration of molecular graphs in drug discovery represents a valuable contribution to the field, and the methodology demonstrates solid technical execution.
The writing is clear and unambiguous. The intro and background show context. All citations are relevant.
Some minor suggestions for improvement:
- The current evaluation could be enriched by incorporating comparisons with additional machine learning models like k-NN, SVM, ERT, XGB, as Logistic Regression is a too simple method. Additionally, deep learning methods can be implemented to provide further support for model's effectiveness.
- A section on Molecular Encoding can be included as there is no detailed information on how a molecule is represented as a graph.
Understanding the black-box model in decision-making is necessary. Please consider using XAI methods or visualizing focused learning regions of feature maps to reveal how the model distinguishes these classes.
The work shows good reproducibility with clear implementation details.
- The manuscript’s novelty is modest, but the focus on molecular graphs for DILI prediction is relevant and timely for drug discovery.
- Please visualize the data preprocessing pipeline (e.g., flowchart/diagram) to improve clarity and reproducibility.
- Replace or complement numeric tables with concise distributional visuals (violin/box plots) to convey variability more effectively.
- Clearly describe the GAT+GCN architecture and its rationale, and illustrate the preprocessing steps that feed the model.
- Strengthen the study by benchmarking against state-of-the-art methods (e.g., TrimNet, AttentiveFP) using identical splits and metrics.
- Provide visual explanations (e.g., attention/attribution maps) to highlight key structures or substructures that drive class discrimination.
- Reproducibility appears adequate. Maintain access to code, data, and detailed settings to enable independent verification.
- Substantiate performance claims through comprehensive comparisons with strong baselines (e.g., TrimNet, AttentiveFP).
- Use violin or box plots across runs/splits to demonstrate variability and robustness of the reported results.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.