Review Reports - Mapping the Chemical Space of Antiviral Peptides with Half-Space Proximal and Metadata Networks Through Interactive Data Mining

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article presents a comprehensive network-based analysis of antiviral peptides, aiming to elucidate the relationships between chemical and biological data. The authors employ Half-Space Proximal Networks and Metadata Networks, an innovative approach that enables high-resolution exploration of chemical space without relying on arbitrary similarity thresholds. Section 2.1 stands out as a significant strength of the article, providing clear explanations of all key concepts, which greatly enhances readability.

A few minor revisions could further strengthen the manuscript:

- The authors predict novel antiviral peptides based on motifs characteristic of known antiviral peptides. However, the current validation predominantly depends on computational methods and existing databases, which may introduce biases. Incorporating experimental validation, even from literature sources, would substantially increase the credibility and impact of the motif discovery.

- The discussion of limitations should be expanded to address potential biases inherent in the utilized databases, as well as the influence of the selected molecular descriptors on the network analysis outcomes.

- The resolution of Scheme 1 and Figure 7 should be improved.

Author Response

Reviewer 1.

General Comment:

A few minor revisions could further strengthen the manuscript:

Comment 1: The authors predict novel antiviral peptides based on motifs characteristic of known antiviral peptides. However, the current validation predominantly depends on computational methods and existing databases, which may introduce biases. Incorporating experimental validation, even from literature sources, would substantially increase the credibility and impact of the motif discovery.

Authors’ Response [AR1]: We thank the reviewer for this important observation. While our validation pipeline was primarily computational, we have strengthened the biological interpretation of the identified motifs by explicitly cross-referencing them with both computationally predicted and experimentally characterized motifs reported in the literature. This comparison revealed convergence with well-established determinants of antiviral activity, thereby adding plausibility to our findings. For example, glycine-initiated motifs have been associated with structural flexibility and binding capacity in SARS-CoV-2 inhibitory peptides, lysine/arginine-rich signatures align with mechanisms of viral envelope destabilization and electrostatic interactions with membranes, and hydrophobic–cationic balances reflect the amphipathic architecture required for membrane disruption. Although these concordances cannot substitute for direct experimental testing, they provide an additional layer of biological support that goes beyond statistical enrichment and highlight that our network-derived motifs are consistent with known antiviral principles.

Manuscript changes: Section 3.4.1 (Motif Enrichment) has been expanded immediately after Table 6 to include a detailed discussion of how representative validated motifs map to established antiviral mechanisms, supported by references.

Comment 2: The discussion of limitations should be expanded to address potential biases inherent in the utilized databases, as well as the influence of the selected molecular descriptors on the network analysis outcomes.

[AR2]: We thank the reviewer for this insightful suggestion. We now explicitly discuss potential sources of bias in motif discovery at the beginning of Section 3.4 (Motif Discovery). In particular, we clarify two limitations: (i) database bias, due to uneven experimental support and heterogeneous assay conditions in AVP annotations and validation sets, and (ii) descriptor/graph dependence, as alternative descriptor sets or similarity metrics could yield different community partitions and motifs. While STREME’s shuffled controls and SEA’s inverse validation help mitigate spurious signals, they cannot fully remove such biases. Accordingly, motif specificity is presented as hypothesis-generating and subject to future experimental validation.

Manuscript changes: A new paragraph has been added at the start of Section 3.4 (Motif Discovery) in red font to highlight these limitations.

Comment 3: The resolution of Scheme 1 and Figure 7 should be improved

[AR3]: We thank the reviewer for this practical observation. Both Scheme 1 and Figure 7 have been substantially improved to enhance visibility and clarity. The updated figures now feature higher resolution and optimized layouts, ensuring that key elements are more easily interpreted by readers

Manuscript changes: Updated versions of Scheme 1 and Figure 7 are now included in the revised manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript presents a technically sophisticated study on mapping the chemical and biological space of antiviral peptides using Half-Space Proximal Networks and Metadata Networks in the StarPep toolbox. The topic is timely and highly relevant given the increasing interest in antiviral peptides as therapeutic agents. The paper is generally well written, methodologically rigorous, and supported by extensive experiments. However, while the work is innovative, several issues limit its clarity, reproducibility, and accessibility. Below, I provide my comments.

1. The manuscript emphasizes the novelty of combining HSPNs with MNs. However, the main contributions of the work are not clearly summarized in the Introduction or Conclusions. I recommend explicitly listing the core contributions at the end of the Introduction.
2. The discussion should better position this approach compared to other peptide network-based methodologies.
3. The description of HSPN construction is detailed but somewhat fragmented. It would benefit from a stepwise workflow diagram summarizing the input, filtering, network generation, thresholding, community detection, scaffold extraction, and motif discovery steps. Moreover, the rationale for choosing Euclidean distance as the sole similarity metric should be more critically discussed, especially since peptide data may benefit from alternative alignment-free metrics. Scaffold extraction procedures (HC vs HB centralities, alignment methods) are extensively described, but the biological interpretation of the selected scaffolds remains limited. Please justify this.
4. Motif discovery yielded 42 motifs, of which 33 were validated. While enrichment analysis is presented, the biological interpretation is relatively brief. The authors should expand the discussion on functional significance, including connections to known antiviral mechanisms or structural motifs. Additionally, the novelty of the motifs (23 reported as new) requires stronger evidence, perhaps with cross-comparison to existing motif libraries.
5. Several figures (network visualizations, heatmaps) are difficult to interpret due to small font sizes and crowded layouts. Improving figure resolution and adding explanatory annotations would enhance readability.
6. While the paper excels in technical network analyses, the biological implications for antiviral peptide discovery and design are underdeveloped. More discussion is needed on: How these network-derived scaffolds/motifs could guide experimental design. Potential therapeutic relevance (e.g., for SARS-CoV-2, HIV, influenza). Limitations of in silico motif validation and the need for experimental confirmation.

In summary, the presented research is interesting and can be accepted after major revision.

Author Response

Reviewer 2.

General Comment:

Comment 1. The manuscript emphasizes the novelty of combining HSPNs with MNs. However, the main contributions of the work are not clearly summarized in the Introduction or Conclusions. I recommend explicitly listing the core contributions at the end of the Introduction.

[AR1]: We thank the reviewer for this valuable suggestion. In the revised manuscript, we now explicitly summarize the core contributions of our work at the end of the Introduction. This addition clearly highlights the novelty of combining Half-Space Proximal Networks (HSPNs) with Metadata Networks (MNs), the extraction of scaffolds and enriched motifs, the validation of their significance, and the practical application of the StarPep toolbox.

Manuscript changes: A new summarizing paragraph has been added in red font at the end of the Introduction (last paragraph), right after the sentence: “Collectively, these findings advance our understanding of AVP structural and functional diversity and their prospects as antiviral therapeutics.”

Comment 2. The discussion should better position this approach compared to other peptide network-based methodologies.

[AR2]: We thank the reviewer for this suggestion. To highlight the novelty of our approach, we have now added a dedicated comparative paragraph in Section 3.2 (immediately after Figure 7). In this addition, we explicitly contrast our framework with previous applications of peptide similarity networks (e.g., tumour-homing, antibiofilm, haemolytic, and antiparasitic peptides [31,34,32,73]. We emphasize three key differentiating elements: (i) the use of HSPNs that can operate without arbitrary similarity cut-offs, yielding high-resolution yet computationally tractable maps of AVP chemical space; (ii) the integration of Metadata Networks (MNs) to provide global biological context (sources, functions, viral targets) that complements the chemical landscape derived from HSPNs; and (iii) the combination of two HSPN topologies (HSPN_OP and HSPN_NC) for scaffold extraction together with a two-stage motif validation pipeline (external enrichment and inverse validation), which strengthens interpretability and supports reuse of scaffold subsets and validated motifs in downstream discovery.

Manuscript changes: A new paragraph in red font has been inserted at the end of Section 3.2, immediately after Figure 7 of the manuscript.

Comment 3. The description of HSPN construction is detailed but somewhat fragmented. It would benefit from a stepwise workflow diagram summarizing the input, filtering, network generation, thresholding, community detection, scaffold extraction, and motif discovery steps. Moreover, the rationale for choosing Euclidean distance as the sole similarity metric should be more critically discussed, especially since peptide data may benefit from alternative alignment-free metrics. Scaffold extraction procedures (HC vs HB centralities, alignment methods) are extensively described, but the biological interpretation of the selected scaffolds remains limited. Please justify this.

[AR3]: We thank the reviewer for this valuable suggestion. As noted, Scheme 1 (now substantially improved during this revision) already provides a stepwise workflow that summarizes the methodology, including data input, filtering, network generation, community detection, scaffold extraction, and motif discovery/enrichment. This scheme was intentionally designed as a compact diagram to balance clarity with the already extensive set of figures included in the manuscript. For readers seeking further technical details on HSPN construction, we refer to our earlier work (Ref. 38 – Aguilera-Mendoza et al., Sci. Rep. 10, 18074, 2020), where Figure 1 presents a detailed schematic of the process. This clarification was included before Figure 3.

Regarding the similarity metric, we acknowledge that different alignment-free distance measures can influence HSPN topology, affecting community distribution and motif detection. In a previous study (Ref. 33 – Castillo-Mendieta et al., npj Syst Biol Appl 10, 115, 2024), we systematically compared Angular Separation, Bhattacharyya, Euclidean, Soergel, and Chebyshev distances using haemolytic peptide networks. That analysis demonstrated that Euclidean, Bhattacharyya, and Soergel produce comparable behaviours in network topology (e.g., cutoff-dependent modularity, degree distributions, scaffold subsets), supporting Euclidean distance as a robust and representative choice. In addition, Euclidean distance (i) is widely used in cheminformatics to compute similarity in multidimensional descriptor spaces, (ii) ensures continuity with our previous peptide network studies, and (iii) yielded stable networks with high topological resolution in exploratory analyses. We have now added a clarifying note in Section 3.2 to explicitly state this rationale and to acknowledge that alternative metrics (e.g., Angular Separation, Chebyshev) may provide complementary perspectives in future work.

Finally, with respect to scaffold extraction, our aim in this study was not to provide exhaustive biological annotation but to generate representative, non-redundant subsets of AVPs that preserve chemical diversity. These scaffold-derived subsets are intended as reusable network resources for downstream applications—such as machine learning training, benchmarking of predictive pipelines, or multi-query similarity searches—rather than as definitive biological classifications. Section 3.3 has been revised to make this rationale explicit and to note that scaffold-level biological interpretation will be pursued in future integrative studies.

Manuscript changes: (i) Scheme 1 has been improved for clarity and readability. (ii) Section 3.2 (opening paragraph, in red font) now clarifies the rationale for choosing Euclidean distance and acknowledges the potential of alternative metrics. (iii) Section 3.3 (second paragraph, in red font) now clarifies the exploratory purpose of scaffold extraction and its role as a reusable resource for downstream applications rather than final biological classifications.

Comment 4. Motif discovery yielded 42 motifs, of which 33 were validated. While enrichment analysis is presented, the biological interpretation is relatively brief. The authors should expand the discussion on functional significance, including connections to known antiviral mechanisms or structural motifs. Additionally, the novelty of the motifs (23 reported as new) requires stronger evidence, perhaps with cross-comparison to existing motif libraries.

[AR4]: We thank the reviewer for this insightful comment. We agree that expanding the biological interpretation of the validated motifs strengthens the contribution of this study. In response, we have expanded Section 3.4 to more explicitly connect the discovered motifs to known antiviral mechanisms and structural signatures. Specifically, we now discuss how glycine-initiated motifs, lysine/arginine-rich patterns, and hydrophobic–cationic balances relate to membrane disruption, viral envelope destabilization, and host–virus protein interactions, referencing recent studies. We also highlight the convergence of several motifs with experimentally validated signatures, such as anti-coronavirus and HIV-inhibitory motifs, further supporting their functional plausibility (Table 6).

To address the reviewer’s point on novelty, we emphasize that out of the 33 validated motifs, only 10 could be cross-referenced with state-of-the-art literature and computational studies integrating curated AVP datasets (e.g., ENNAVIA-D, anti-coronavirus peptide models). The remaining 23 motifs appear to be novel, as they were not found in recent literature sources and showed no matches in PROSITE’s pattern and profile databases when queried with the Motif Search engine of the GenomeNet suite. These negative results reinforce the interpretation that the 23 validated motifs represent previously undescribed antiviral sequence signatures. We also clarify in the revised text that, while computational validation provides strong support, their biological functionality will ultimately require experimental confirmation.

Together, these additions reinforce the biological significance of the motifs and provide stronger justification for their novelty.

Manuscript changes: Section 3.4.1 (Motif Discovery) has been expanded to include a mechanistic discussion that links representative motifs (e.g., glycine-initiated, lysine/arginine-rich, and hydrophobic–cationic patterns) to established antiviral mechanisms and structural determinants, with appropriate references. The title of Table 6 has been revised, and a clarifying footnote was added to enhance its self-explanatory value. Furthermore, the final paragraph of Section 3.4.1 now specifies that 23 motifs could not be traced in recent literature or matched to PROSITE’s pattern/profile databases, thereby supporting their novelty while also noting that experimental validation will be necessary to confirm their biological functionality.

Comment 5. Several figures (network visualizations, heatmaps) are difficult to interpret due to small font sizes and crowded layouts. Improving figure resolution and adding explanatory annotations would enhance readability.

[AR5]: We thank the reviewer for this valuable observation. In response, we have substantially improved the resolution, readability, and annotations of the figures highlighted by the reviewer. Specifically, Scheme 1 and Figures 1, 2, 4A, 4B, and 7 were revised with higher-resolution graphics, enlarged font sizes, and clearer layouts. Additional explanatory annotations were also incorporated where needed to facilitate interpretation. We believe these updates significantly enhance figure clarity and overall readability.

Manuscript changes: Scheme 1 and Figures 1, 2, 4A, 4B, and 7 have been replaced with revised, high-resolution versions including larger fonts and improved annotations.

Comment 6. While the paper excels in technical network analyses, the biological implications for antiviral peptide discovery and design are underdeveloped. More discussion is needed on: How these network-derived scaffolds/motifs could guide experimental design. Potential therapeutic relevance (e.g., for SARS-CoV-2, HIV, influenza). Limitations of in silico motif validation and the need for experimental confirmation.

[AR6]: We thank the reviewer for this important comment. In the revised manuscript, we have strengthened the discussion of the biological implications of our network-derived scaffolds and motifs. Specifically, in Section 3.3, we now highlight how scaffold subsets, while primarily intended as non-redundant datasets for modelling and benchmarking, may also serve as rational entry points for experimental screening panels, helping to prioritize chemically diverse templates for synthesis and in vitro testing. In Section 3.4.1 we expanded the discussion to clarify how short, validated motifs could be embedded into synthetic constructs or grafted onto scaffolds to guide therapeutic peptide design against relevant viral targets, including HIV, influenza, and SARS-CoV-2. Finally, in Section 3.4.2, we added a dedicated statement acknowledging the limitations of in silico validation, emphasizing that these motifs remain computationally validated and should be regarded as hypothesis-generating elements, with experimental confirmation required to establish antiviral functionality.

Manuscript changes: In Section 3.3, a new opening paragraph clarifies the primary role of scaffolds as chemically diverse, representative entry points for experimental panels, while the closing paragraph now outlines their potential use in the discovery of therapeutic candidates. In Section 3.4.1, the final two paragraphs were expanded to describe how validated motifs can inform peptide design strategies and be tailored toward therapeutically relevant viral targets. In Section 3.4.2, the concluding paragraphs were revised to emphasize the limitations of in silico motif validation and the need for subsequent experimental confirmation.

Comment 7: In summary, the presented research is interesting and can be accepted after major revision.

[AR7]: We thank the reviewer for the constructive summary. All major revisions requested have been carefully addressed in the revised manuscript, including expanded biological interpretation of scaffolds and motifs, improved figure clarity, explicit discussion of therapeutic relevance and limitations, and a stronger justification of motif novelty. We believe these additions substantially strengthen the manuscript and align it with the reviewer’s recommendations

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript presents an extensive chemoinformatics and network-analysis study of antiviral peptides (AVPs) using Half-Space Proximal Networks (HSPNs), Metadata Networks (MNs), and the StarPep toolbox. It provides a novel integration of chemical and biological information, motif discovery, and scaffold extraction strategies. Overall, the manuscript can be accepted after minor revision.

Comments：

Although motifs are statistically validated, experimental or literature-based biological validation is limited. A deeper discussion on how identified motifs align with known structural determinants of AVPs (beyond citing overlaps with a few studies) would strengthen the biological impact.
Network figures (HSPNs, MNs) are somewhat difficult to interpret due to high density. Enlarged or simplified subgraphs with clear labeling would help readers appreciate the structural insights.

Author Response

Reviewer 3.

General Comment:

Comment 1: Although motifs are statistically validated, experimental or literature-based biological validation is limited. A deeper discussion on how identified motifs align with known structural determinants of AVPs (beyond citing overlaps with a few studies) would strengthen the biological impact.

[AR1]: We thank the reviewer for this observation. As also suggested by Reviewer 2, we expanded Section 3.4.1 to explicitly connect validated motifs with known antiviral mechanisms and structural determinants, including glycine-initiated flexibility, lysine/arginine-rich charge interactions, and hydrophobic–cationic balances relevant for membrane disruption and envelope destabilization. We also added a paragraph acknowledging the limitations of purely in silico validation and emphasizing the need for experimental confirmation.

Manuscript changes: In Section 3.4.1 (final two paragraphs), we expanded the discussion to more clearly connect the validated motifs with known structural determinants of antiviral peptides. Specifically, we now describe how glycine-initiated motifs contribute to structural flexibility, how lysine/arginine-rich patterns promote electrostatic interactions with viral envelopes, and how hydrophobic–cationic balances enable membrane disruption and viral destabilization. In addition, we added a new paragraph at the end of Section 3.4.2 explicitly acknowledging the limitations of in silico validation and highlighting the need for experimental confirmation of motif functionality.

Comment 2: Network figures (HSPNs, MNs) are somewhat difficult to interpret due to high density. Enlarged or simplified subgraphs with clear labeling would help readers appreciate the structural insights.

[AR2]: We agree with this comment. Following similar feedback from Reviewer 2, we improved Scheme 1 and Figures 1, 2, 4A, 4B, and 7 by increasing resolution, enlarging fonts, reducing crowding, and adding annotations to enhance readability and interpretation. Please see the revised version of the manuscript

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Thanks, I have no additional questions