Molecular Modelling of Anti-Inflammatory Activity: Application of the ToSS-MoDE Approach to Synthetic and Natural Compounds

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper investigates anti-inflammatory activity QSAR classification model based on ToSS-MoDE framework (weighted spectral moments created in MODESLAB) and Linear Discriminant Analysis (LDA). The subject is timely and the concept of utilizing graph/spectral descriptors as a filtering method in synthetic molecules and in plant metabolites is interesting. With that said, the paper, as presented, is currently lacking in terms of reproducibility and even the simplest transparency of the dataset, and some of the conclusions that it makes are even stronger than the presented work supports.

Major Comments

• This manuscript describes a training set of 410 compounds (180 active and 230 inactive) and an external validation set consisting of 62 compounds, but there is no clear description of the criteria used to categorize active and inactive (endpoint/assay type, thresholds, and data sources). To facilitate reproducibility, we would like to have the identifiers of the activities labels used by the compounds and the source information displayed, the MODESLAB descriptor-generation settings/workflow should be recorded, and the final discriminant equation and classification rule should be presented in a clean, unambiguous form (this information could be included in Supplementary Materials).
• Reporting on performance is not complete (it is not enough to be accurate). Generally, the paper indicates that classification is correct (91.59% training; 90.2% validation) but a classifier must also provide confusion matrices and standard measures (sensitivity/specificity, balanced accuracy, MCC) to prevent false impressions, in particular, the presence of class imbalance and varying costs of false positives vs false negatives.
• External validation processing requires clarification, in particular, the cases of the unclassified one. The reported 62 compounds are referred to as the external set, however, 11 of these are unclassified, and the 90.2% appears to be calculated on the remaining 51 compounds.
• Claims of experimental validation of diosgenin are exaggerated and self-contradictory. The authors are categorical in the discussion that, the positive result of a plant extract cannot enable them to conclude that diosgenin is anti-inflammatory, as other metabolites exist. Nevertheless, the conclusion goes on to say that diosgenin has been experimentally validated and strengthens its promise. Recommendation: This conclusion should be made more modest. Depending on the presented information, diosgenin at least seems to be a theoretical forecast (with stated probability) and not experimentally confirmed throughout the text.
• It seems that some entries are repeated in tables which should be purged and transferred to Supplementary Materials. The training / prediction tables are also very long and contain blatant repetitions (e.g., the same nifuresimizone, Intrazole/Isonixin/Isopenax/isoprazone).

  Minor remarks and editorial problems.

  Administrative statements near the end need editing:

  • There should be administrative statements that are made close to the end and should be edited.
  • Institutional Review Board Statement:
  • Not aplicable. → Not applicable.
  • Informed Consent Statement: Not aplicable. → Not applicable.
  • Acknowledge (If: applicable). - Acknowledgments (if applicable).

   the paper states spectral moments µ₀–µ₁₅ are used and weighted by bond properties, but the weighting scheme and any preprocessing (scaling/standardization) are not clearly described. A short clarification would help.

Author Response

Response to Reviewers

Manuscript ID: biophysica-4111845
Title: Molecular Modelling of Anti-inflammatory Activity: Application of the ToSS-MoDE Approach to Synthetic and Natural Compounds

We sincerely thank the reviewer for their thorough evaluation of our manuscript and for the constructive comments, which have helped us to significantly improve the scientific quality, clarity, and reproducibility of the work. All comments have been carefully addressed below. Changes made to the manuscript are highlighted in yellow in the revised version.

 

Reviewer 1 – Major Comments

 

Comment 1

The manuscript describes a training set of 410 compounds and an external validation set of 62 compounds, but there is no clear description of the criteria used to categorize compounds as active or inactive. Identifiers, data sources, descriptor-generation settings, and the final discriminant equation should be clearly reported to ensure reproducibility.

Response:
We thank the reviewer for highlighting this important issue. We fully agree that transparency and reproducibility are essential, particularly for QSAR studies.

To address this concern, the manuscript has been substantially revised:

• A detailed description of the dataset composition and activity labeling criteria has been added to Section 2.1 (Dataset).
• For each compound, chemical identifiers (compound name, PubChem CID, InChIKey, SMILES), data sources, assay type, experimental endpoint, activity thresholds, and bibliographic references are now provided in Supplementary Table S1.
• Activity classification is now explicitly described as being based on experimentally reported anti-inflammatory assays (e.g., COX inhibition, suppression of inflammatory mediators, and in vivo inflammation models), using thresholds defined in the original literature.
• The descriptor-generation workflow in MODESLAB, including weighting schemes and preprocessing steps, is fully detailed in Section 2.2 and Supplementary Table S2.
• The final discriminant equation, coefficients, centroids, and classification rules are now explicitly reported in Section 2.3 and Supplementary Table S3.

These additions ensure full reproducibility of the proposed QSAR model.

 

Comment 2

Reporting of model performance is incomplete. Accuracy alone is insufficient; confusion matrices and additional metrics such as sensitivity, specificity, balanced accuracy, and MCC are required.

Response:
We agree with the reviewer and appreciate this suggestion.

The performance evaluation has been expanded as follows:

• Confusion matrices for the training and external validation sets are now reported and discussed.
• In addition to overall accuracy, we now report sensitivity, specificity, balanced accuracy, and Matthews correlation coefficient (MCC) for the training set (Section 3.1).
• The MCC value obtained (0.83) indicates strong and balanced discrimination between active and inactive compounds.
• These metrics are now explicitly discussed to avoid misleading interpretations, particularly in the presence of class imbalance.

 

Comment 3

Clarification is required regarding the external validation set, particularly the handling of unclassified compounds. The reported 90.2% accuracy appears to be calculated only for 51 of the 62 compounds.

Response:
We thank the reviewer for this observation and have clarified this point in the revised manuscript.

• As now explicitly stated in Section 3.2 (External Validation), 11 compounds were labeled as unclassified due to posterior probabilities below 0.60.
• The reported 90.2% accuracy refers exclusively to the 51 compounds that were confidently classified.
• To avoid optimistic bias, we now additionally report a conservative accuracy of 74.2%, obtained by considering unclassified compounds as misclassified.
• Both values are now reported and discussed to provide a transparent and balanced evaluation of model performance.

 

Comment 4

Claims regarding experimental validation of diosgenin are exaggerated and internally inconsistent. The conclusions should be more cautious.

Response:
We fully agree and thank the reviewer for pointing out this inconsistency.

The manuscript has been revised to ensure a cautious and consistent interpretation:

• All statements implying experimental validation of isolated diosgenin have been removed.
• Diosgenin is now consistently described as a QSAR-based theoretical prediction, supported by a posterior probability of 0.81.
• We clearly state that anti-inflammatory activity observed in plant extracts cannot be attributed to diosgenin alone, as multiple metabolites are present.
• The Discussion and Conclusion sections now explicitly emphasize that experimental validation of isolated diosgenin is required.

 

Comment 5

Some compounds are repeated in tables, and the training/prediction tables are excessively long.

Response:
We appreciate this comment.

• All duplicate compound entries have been carefully reviewed and removed.
• Long tables containing posterior probabilities and classification details have been moved to the Supplementary Materials, while the main text now presents summarized results and representative examples.
• This restructuring improves readability without compromising data availability.

 

Reviewer 1 – Minor Comments and Editorial Issues

 

Comment 6

Administrative statements require editing, and the descriptor weighting and preprocessing procedures should be clarified.

Response:
All editorial corrections have been implemented:

• “Not aplicable” has been corrected to “Not applicable”.
• “Acknowledge” has been corrected to “Acknowledgments”.
• Administrative statements have been edited for consistency and clarity.

Additionally, a concise clarification of the spectral moment weighting schemes (µ₀–µ₁₅) and descriptor preprocessing (z-score standardization) has been added to Section 2.2, ensuring methodological transparency.

 

Final Statement

Once again, we thank the reviewer for their valuable and constructive feedback. We believe that the revisions have significantly improved the manuscript’s transparency, reproducibility, and scientific rigor, and we hope that the revised version is now suitable for publication in Biophysica.

 

Reviewer 2 Report

Comments and Suggestions for Authors

Dear colleagues, thank you for your research and attention to such an important problem as computer assisted design of new anti-inflammatory drugs. Although the topic is still of a great importance, there have been many research reporting a vast amount of new compounds with potential anti-inflammatory activity waiting for their properties to be confirmed. Moreover, there are many alternative ways to predict biological activities of chemical compounds rather than QSAR. Unfortunately, the manuscript contains no information on the superiority of the spectral moments based QSAR approach over alternatives. 

The Introduction section is rather brief, I would say, too brief to evaluate the significance of the research topic, practical and scientific importance as well as novelty of the research.  

Section 3. Results and discussion contains two very long tables (Table 1 and Table 2) that could be placed into a supplementary section in more convenient format (spreadsheets, for example). Also titles of the tables are way too laconical providing essentially no information to the context. 

The phrase "One of the most important criteria for accepting or rejecting a discriminant model, as shown in this work, is based on the statistics of the external prediction series." is slightly unfair towards colleagues of us. This is a commonly agreed criteria to confirm quality of statistical models that has been suggested long before this work. 

One more remark is about figures and their titles: the pictures are blurred and the titles add no any useful explanation of the content. For example, "Figure 2. Partitioning of anti-inflammatory activity. Regions of isocontribution." There are some numbers, but their meaning is completely unclear. In text: "As we can see in the diagram, the isoactive zones in the benzene and furan rings contribute 3.72 and 2.37 respectively". It would be better to add some clarification of the calculated values and units throughout the manuscript. 
Unfortunately, I can not recommend the manuscript in present form for publication.

Author Response

Response to Reviewer 2

Manuscript ID: biophysica-4111845
Title: Molecular Modelling of Anti-inflammatory Activity: Application of the ToSS-MoDE Approach to Synthetic and Natural Compounds

We sincerely thank the reviewer for the thoughtful and constructive comments. We acknowledge the concerns raised and have revised the manuscript extensively to improve the contextualization, clarity, presentation quality, and positioning of the proposed QSAR methodology. All modifications are highlighted in yellow in the revised manuscript.

 

General Comment

Although the topic remains important, many alternative approaches exist for predicting biological activity, and the manuscript does not explain the advantages of spectral moment–based QSAR models over other methods.

Response:
We appreciate this important remark and agree that the original version of the manuscript did not sufficiently contextualize the advantages of the proposed approach.

To address this point, the Introduction has been significantly expanded to clearly position the ToSS-MoDE/spectral moment–based QSAR framework relative to alternative methodologies such as molecular docking, molecular dynamics, pharmacophore modeling, machine learning, and deep learning approaches.

Specifically, we now explicitly discuss that:

• Spectral moment–based QSAR models offer low computational cost, high interpretability, and robust performance on large and chemically diverse datasets.
• Unlike structure-based methods, this approach does not require three-dimensional optimization or target structure availability, making it particularly suitable for early-stage virtual screening.
• The ToSS-MoDE methodology enables direct interpretation of substructural contributions to activity, which complements black-box machine learning approaches.

These points are now detailed in the final paragraphs of the Introduction, with appropriate recent references added.

 

Comment on the Introduction Section

The Introduction section is too brief to evaluate the significance, practical importance, and novelty of the study.

Response:
We fully agree with this assessment.

The Introduction has been substantially revised and expanded to:

• Provide a clearer overview of the current challenges in anti-inflammatory drug discovery.
• Highlight recent advances in computational and chemoinformatics-based approaches.
• Clearly articulate the novelty of applying the ToSS-MoDE framework to a mixed dataset of synthetic compounds and natural metabolites, including medicinal plant constituents.
• Clarify the practical relevance of the method for high-throughput screening and rational drug design.

This revision significantly strengthens the scientific motivation and novelty of the work.

 

Comment on Tables in Section 3 (Results and Discussion)

Tables 1 and 2 are excessively long and should be moved to the Supplementary Materials. Table titles are too laconic and lack contextual information.

Response:
We thank the reviewer for this suggestion and have revised the manuscript accordingly.

• Tables 1 and 2 have been moved to the Supplementary Materials, where they are now provided in a clearer and more convenient format.
• In the main manuscript, only summarized results and representative examples are now discussed.
• All table titles have been rewritten to be fully descriptive, explicitly stating the contents (compound set, classification type, posterior probabilities, and experimental labels).

These changes improve readability and presentation quality.

 

Comment on the Statement Regarding External Validation

The statement suggesting that external validation statistics are one of the most important criteria for model acceptance appears unfair, as this criterion is well established prior to this work.

Response:
We appreciate this observation and agree with the reviewer.

The sentence has been rephrased to acknowledge that:

• External validation is a widely accepted and long-established criterion in QSAR and statistical modeling.
• Our work follows, rather than introduces, this established best practice.

The revised text now appropriately credits the broader scientific community and avoids any implication of originality regarding this criterion.

 

Comment on Figures and Figure Captions

Figures are blurred, captions are uninformative, and numerical values shown in figures are not properly explained. Units and meaning of calculated values are unclear.

Response:
We thank the reviewer for pointing out these important presentation issues.

The manuscript has been revised as follows:

• All figures have been replaced with higher-resolution versions to improve readability.
• Figure captions have been substantially expanded, now providing clear explanations of:
• What the highlighted regions represent,
• The meaning of numerical values shown in the figures,
• How these values relate to bond contributions and spectral moments.
• In the main text, we now explicitly clarify that the reported numerical values (e.g., 3.72, 2.37) represent relative contributions to the discriminant function, derived from the weighted spectral moment terms, and are dimensionless model coefficients, not physical units.

These clarifications improve the interpretability of both the figures and the associated discussion.

 

Final Comment

The manuscript cannot be recommended for publication in its present form.

Response:
We thank the reviewer for their honest and constructive assessment. In response to these concerns, the manuscript has undergone major revision, including expansion of the Introduction, improved positioning of the methodology, reorganization of tables, correction of statements, and substantial enhancement of figure quality and explanations.

We believe that these changes significantly improve the manuscript’s scientific clarity, context, and presentation quality, and we sincerely hope that the revised version will now be suitable for publication in Biophysica.

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for your revised version of the manuscript that has been improved significantly and now I have no questions and would recommend the manuscript for publication.