Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I was glad to review this manuscript entitled "AI-Driven Design of High Affinity Biomolecule-Drug Conjugates for Gynecological Cancer Therapy". This narrative review summarizes the innovative practice of artificial intelligence (AI) in developing high-affinity biomolecule-drug conjugates to treat gynecological cancer and concludes that AI is a powerful tool for advancing precision oncology through rational, data-driven, and personalized therapeutic design. The topic is very interesting. The manuscript can be accepted for publication pending some minor corrections:

 1) The title of this manuscript could be "AI-Driven Design of High Affinity Biomolecule-Drug Conjugates for Gynecological Cancer Therapy: An Up-to-Date Narrative Review".

2) I would suggest making the abstract structured, including all parts (background, aims, etc.)

3) Please include the primary and secondary aims of the study at the end of the introduction section

4) The abbreviations are not explained in Table 1.

5) I would suggest adding a figure for section 3.3. Binding Kinetics and Affinity Landscapes Modeling

6) I would also suggest adding a figure for 3.6. Human-Centred Interpretation of AI Predictions

7) Briefly describe the differences between volumetric arc therapy and three-dimensional conformal therapy for gynecological cancer. 

8) The medical industry has seen fast and ongoing technical advancements in recent years. The IoT concept's adoption into medical practice was one of the most revolutionary developments. Give a succinct explanation of how IOT affects gynecological cancers.

9) Finally, briefly describe the role of bevacizumab in gynecological cancers. 

Author Response

Reviewer # 1
I was glad to review this manuscript entitled "AI-Driven Design of High Affinity Biomolecule-Drug Conjugates for Gynecological Cancer Therapy". This narrative review summarizes the innovative practice of artificial intelligence (AI) in developing high-affinity biomolecule-drug conjugates to treat gynecological cancer and concludes that AI is a powerful tool for advancing precision oncology through rational, data-driven, and personalized therapeutic design. The topic is very interesting. The manuscript can be accepted for publication pending some minor corrections:

Response:    We sincerely thank the reviewer for the careful evaluation of our manuscript and for the encouraging and constructive comments. We are grateful for the reviewer’s positive assessment of the novelty and relevance of our review on AI-driven biomolecule–drug conjugate design in gynecological cancer. All the suggested revisions and minor corrections have been carefully addressed in the revised manuscript to improve its scientific clarity, depth, and overall presentation. Detailed point-by-point responses are provided below.

1)    The title of this manuscript could be "AI-Driven Design of High Affinity Biomolecule-Drug Conjugates for Gynecological Cancer Therapy: An Up-to-Date Narrative Review".

Response:    We thank the reviewer for this valuable suggestion. As recommended, the title of the manuscript has been revised to better reflect the scope and contemporary relevance of the review. The revised title now reads: “AI-Driven Design of High Affinity Biomolecule-Drug Conjugates for Gynecological Cancer Therapy: An Up-to-Date Narrative Review.” This modification has been incorporated in the revised manuscript.

2)    I would suggest making the abstract structured, including all parts (background, aims, etc.).

Response:    We thank the reviewer for this constructive suggestion. As recommended, the abstract has been thoroughly revised into a structured format to improve clarity, organization, and readability. The revised abstract now includes distinct sections covering the background, objective, key discussion areas, major findings, and conclusion of the review. Additionally, the language has been refined to improve scientific clarity and reduce redundancy. The structured abstract has been incorporated into the revised manuscript.

3)    Please include the primary and secondary aims of the study at the end of the introduction section.

Response:    We thank the reviewer for this helpful suggestion. As recommended, the primary and secondary aims of the review have now been clearly stated at the end of the introduction section to better define the scope and objectives of the manuscript. This addition improves the overall organization and conceptual clarity of the review. The following text has been incorporated into the revised manuscript:

4)    The abbreviations are not explained in Table 1.

Response:    We thank the reviewer for identifying this oversight. As suggested, all abbreviations used in Table 1 have now been clearly defined as footnote/legend in the table to improve readability and ensure clarity for readers from multidisciplinary backgrounds. The Table 1 has been updated accordingly in the revised manuscript.
5)    I would suggest adding a figure for section 3.3. Binding Kinetics and Affinity Landscapes Modeling.

Response:    We sincerely thank the reviewer for this valuable suggestion. In response, a new illustrative as Figure 3, has been incorporated in Section 3.3 to provide a clearer conceptual understanding of AI-assisted binding kinetics and affinity landscape modelling in biomolecule–drug conjugate design. The newly added figure schematically illustrates the integration of AI-driven predictive frameworks with biomolecular interaction dynamics, including association/dissociation kinetics, affinity optimization, sequence–structure relationships, and therapeutic targeting in heterogeneous gynecological tumors. This addition improves the visual interpretation and conceptual clarity of the section.

6)    I would also suggest adding a figure for 3.6. Human-Centered Interpretation of AI Predictions.

Response:    We thank the reviewer for this insightful recommendation. Accordingly, a new schematic figure as Figure 4, has been added to Section 3.6 to visually summarize the concept of explainable and human-centered AI in biomolecule–drug conjugate design. The figure highlights how explainable AI frameworks improve interpretability, transparency, and clinical trust by identifying key molecular determinants influencing AI predictions. This addition strengthens the translational perspective of the manuscript and improves reader comprehension of explainable AI approaches in precision oncology.

7)    Briefly describe the differences between volumetric arc therapy and three-
dimensional conformal therapy for gynecological cancer. 

Response:    We thank the reviewer for this valuable suggestion. As required, a brief comparative description of volumetric modulated arc therapy (VMAT) and three-dimensional conformal radiotherapy (3D-CRT) has been added in the revised manuscript under section of introduction. This addition clarifies the evolution of radiotherapy techniques and their relevance in improving treatment precision for gynecological cancers.

8)    The medical industry has seen fast and ongoing technical advancements in recent years. The IoT concept
's adoption into medical practice was one of the most revolutionary developments. Give a succinct 
explanation of how IOT affects gynecological cancers.

Response:    We thank the reviewer for this insightful comment. A concise description of the role of Internet of Things (IoT) in gynecological cancer management has been incorporated in the revised manuscript under subsection 8.1 (Future Prospective), highlighting its relevance in remote monitoring, personalized care, and data-driven oncology.

9)    Finally, briefly describe the role of bevacizumab in gynecological cancers. 

Response:    We sincerely thank the reviewer for this suggestion. A brief overview of bevacizumab and its clinical role in gynecological cancers have been added to the introduction section of the revised manuscript to strengthen the discussion on targeted anti-angiogenic therapies.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript by Garg and co-authors presents a comprehensive review of artificial intelligence applications in the design and optimization of biomolecule-drug conjugates for gynecological cancer therapy. The authors discuss AI-driven approaches for predicting binding affinity, structural compatibility, linker stability, payload selection, intracellular trafficking, and resistance mechanisms. While the topic is timely and important, several significant issues need to be addressed before publication.

(1) The manuscript attempts to cover a very wide range of topics, from multi-omics integration to linker chemistry to regulatory considerations, but fails to provide sufficient depth or critical evaluation in any single area. Many claims about AI capabilities (e.g., "AI models can predict with high precision...") are stated as facts without supporting data or citation of specific validated models in gynecological cancer area. It is desirable to narrow the scope or restructure to focus on 2-3 areas where AI has demonstrated sensible progress, with critical discussion and limitations.

(2) While the title is focused on gynecological cancers, the manuscript mainly discusses general principles of AI-driven conjugate design that would apply equally to any cancer type. Specific examples of ovarian, cervical, or endometrial cancer applications are sparse. The authors cite only several gynecological cancer-specific studies (e.g., references 2, 17, 22, 35-36), and these are not discussed in meaningful detail.  Please strengthen the discussion of gynecological  challenges (e.g., ovarian cancer heterogeneity, cervical cancer HPV-related targets) and how AI approaches address them.

(3) While the manuscript claims that AI approaches overcome limitations of "empirical" methods, it does not provide systematic comparisons showing improvements (reduced development time, improved prediction accuracy, higher clinical success rates). Without such benchmarks, the superiority of AI-driven design remains unsupported.

(4) The manuscript uses "gynecological" and "gynaecological" terms. Choose unique spelling.

Summarizing, I recommend major revision of the manuscript before acceptance.

Author Response

Reviewer # 2
The manuscript by Garg and co-authors presents a comprehensive review of artificial intelligence applications in the design and optimization of biomolecule-drug conjugates for gynecological cancer therapy. The authors discuss AI-driven approaches for predicting binding affinity, structural compatibility, linker stability, payload selection, intracellular trafficking, and resistance mechanisms. While the topic is timely and important, several significant issues need to be addressed before publication.

Response:    We sincerely thank the reviewer for the careful evaluation of our manuscript and for the constructive comments provided. We appreciate the reviewer’s recognition of the relevance and timeliness of the topic, particularly the emerging role of AI in biomolecule-drug conjugate engineering for gynecological cancers. We acknowledge the concerns raised regarding scientific depth, specificity, methodological clarity, and translational perspectives. In response, the manuscript has been extensively revised to improve its overall scientific rigor, conceptual clarity, gynecological cancer-specific focus, and critical discussion of current limitations and future challenges. Additional explanatory text, revised sections, updated figures, and expanded discussions on AI methodologies, translational barriers, and clinical applicability have been incorporated throughout the manuscript. Detailed point-by-point responses to all reviewer comments are provided below.

1)    The manuscript attempts to cover a very wide range of topics, from multi-omics integration to linker chemistry to regulatory considerations but fails to provide sufficient depth or critical evaluation in any single area. Many claims about AI capabilities (e.g., "AI models can predict with high precision...") are stated as facts without supporting data or citation of specific validated models in gynecological cancer area. It is desirable to narrow the scope or restructure to focus on 2-3 areas where AI has demonstrated sensible progress, with critical discussion and limitations.

Response:    We sincerely thank the reviewer for this important and insightful comment. We acknowledge that the original manuscript covered a broad range of AI-assisted biomolecule-drug conjugate design topics with limited critical depth in certain areas. In response, the manuscript has been substantially revised to improve its scientific focus, analytical depth, and translational perspective. The revised manuscript now places greater emphasis on key areas where AI-assisted approaches have demonstrated comparatively meaningful progress in gynecological cancer-related conjugate engineering, particularly: (i) biomolecule–target binding affinity and structural interaction prediction, (ii) linker and payload optimization, and (iii) intracellular trafficking and resistance-associated therapeutic response modelling. 

To address the reviewer’s concern regarding overgeneralization of AI capabilities, generalized statements throughout the manuscript have been revised to adopt a more evidence-based and balanced tone. Assertions implying definitive predictive capability have been replaced with more cautious phrasing such as “AI-assisted frameworks have demonstrated promising predictive potential” and “emerging studies suggest possible utility in optimization and therapeutic modelling.” In addition, Section 2.6 has been substantially expanded to include a focused discussion on current translational limitations, including restricted gynecological cancer-specific datasets, dataset heterogeneity, model interpretability challenges, limited experimental and prospective clinical validation, reproducibility concerns, and constraints affecting clinical generalizability. Additional methodological clarification, representative examples, and critical discussion of current limitations have also been incorporated throughout the revised manuscript. Collectively, these revisions provide a more balanced, critically evaluated, and translationally grounded perspective on the current status and future potential of AI-guided biomolecule–drug conjugate engineering in gynecological oncology.

2)    While the title is focused on gynecological cancers, the manuscript mainly discusses general principles of AI-driven conjugate design that would apply equally to any cancer type. Specific examples of ovarian, cervical, or endometrial cancer applications are sparse. The authors cite only several gynecological cancer-specific studies (e.g., references 2, 17, 22, 35-36), and these are not discussed in meaningful detail.  Please strengthen the discussion of gynecological challenges (e.g., ovarian cancer heterogeneity, cervical cancer HPV-related targets) and how AI approaches address them.

Response:    We sincerely thank the reviewer for this valuable and constructive observation. We acknowledge that the initial version of the manuscript discussed several AI-driven conjugate engineering principles in a relatively generalized manner, which limited the emphasis on gynecological cancer-specific applications and challenges. In response, the manuscript has been substantially revised to strengthen its disease-focused perspective and improve the integration of ovarian, cervical, and endometrial cancer-specific examples throughout the discussion. Specifically, additional discussion has been incorporated addressing key biological and therapeutic challenges associated with gynecological malignancies, including tumor heterogeneity and genomic instability in ovarian cancer, HPV-associated oncogenic signaling and immune evasion in cervical cancer, and hormonally regulated molecular progression in endometrial cancer. The revised manuscript now more clearly highlights how AI-assisted approaches may support context-specific target identification, subtype-specific affinity prediction, linker and payload optimization, and resistance-aware therapeutic modelling in these disease settings.

Furthermore, gynecological cancer-relevant targets and biomarkers, including FRα, HER2, Trop-2, and HPV-associated molecular markers, are now discussed in greater mechanistic and translational detail (Section 3). Previously cited gynecological cancer-specific studies have also been more critically integrated into the manuscript with expanded discussion of their methodological significance and clinical relevance rather than simple citation alone. Collectively, these revisions improve the alignment between the manuscript title, scientific scope, and gynecological oncology focus, thereby strengthening the translational relevance and disease specificity of the review.

3)    While the manuscript claims that AI approaches overcome limitations of "empirical" methods, it does not provide systematic comparisons showing improvements (reduced development time, improved prediction accuracy, higher clinical success rates). Without such benchmarks, the superiority of AI-driven design remains unsupported.

Response:    We sincerely thank the reviewer for this important and insightful comment. We agree that claims regarding the superiority of AI-assisted conjugate engineering should be presented with appropriate scientific balance and supported through critical comparative discussion rather than generalized assertions. In response, the manuscript has been revised to include a more balanced comparison between conventional empirical approaches and emerging AI-assisted biomolecule–drug conjugate design strategies. Specifically, additional discussion has been incorporated highlighting the potential advantages of AI-assisted frameworks in accelerating virtual screening, improving biomolecule-target interaction prediction, optimizing linker and payload selection, and reducing dependence on repetitive trial-and-error experimental workflows.

At the same time, the revised manuscript now clearly emphasizes that many currently available AI-assisted models remain under limited experimental, preclinical, or early translational validation. The manuscript further acknowledges that definitive benchmarking demonstrates reduced development timelines, consistently improved predictive accuracy, or enhanced clinical success rates in gynecological oncology settings remains limited and continues to evolve. To improve scientific balance, several generalized statements implying definitive superiority of AI-based methods have been revised throughout the manuscript and replaced with more cautious, evidence-based descriptions. In addition, an expanded comparative discussion of conventional therapies, traditional conjugate engineering strategies, and AI-guided biomolecule–drug conjugate development has been incorporated and summarized in the revised Table 3. Collectively, these revisions provide a more critically grounded and translationally balanced discussion of the current capabilities, limitations, and future potential of AI-assisted conjugate engineering in gynecological cancers.

4)    The manuscript uses "gynecological" and "gynaecological" terms. Choose unique spelling.
Summarizing, I recommend major revision of the manuscript before acceptance.

Response:    We sincerely thank the reviewer for this careful observation. The manuscript has been thoroughly revised to ensure consistency in terminology and spelling throughout the text. Specifically, the spelling has now been standardized uniformly to “gynecological” across the entire manuscript, including the title, abstract, main text, tables, and figure legends. In addition, the manuscript has undergone careful language editing to improve overall consistency, clarity, and readability.

 

Reviewer 3 Report

Comments and Suggestions for Authors

This perspective addresses a timely and important topic at the intersection of artificial intelligence (AI), biomolecule–drug conjugates, and gynecological oncology. The manuscript provides a broad conceptual overview of how AI may facilitate affinity prediction, linker optimization, payload engineering, intracellular trafficking analysis, and resistance prediction in next-generation conjugate therapeutics. The topic is highly relevant given the increasing interest in AI-assisted precision oncology.

However, despite its conceptual appeal, the manuscript currently remains largely descriptive and speculative. Many sections summarize general expectations of AI without providing sufficient technical depth, mechanistic insight, critical evaluation, or translational evidence. The article frequently uses broad statements regarding the transformative capability of AI, yet lacks detailed discussion of actual implemented workflows, validated case studies, model limitations, benchmarking strategies, and regulatory feasibility. As a result, the manuscript reads more as a generalized narrative overview rather than a rigorous perspective grounded in current technological realities.

Substantial revision is recommended before the manuscript can be considered for publication.

 

Major Comments

1. The manuscript lacks technical depth regarding AI methodologies

The manuscript repeatedly refers to “AI,” “machine learning,” and “deep learning” as transformative tools, but the discussion remains highly generic throughout most sections. For example, CNNs, transformers, GNNs, and transfer learning are mentioned, yet the manuscript does not adequately explain:

• what specific inputs are used,
• what outputs are predicted,
• how models are trained,
• what datasets are required,
• what performance metrics are used,
• or how uncertainty is quantified.

The article would benefit substantially from concrete examples of implemented AI frameworks used in antibody-drug conjugate (ADC) or biomolecule–drug conjugate optimization.

For instance:

• Which GNN architectures are most suitable for affinity prediction?
• How are structural embeddings generated?
• Are AlphaFold-derived structures sufficiently accurate for conjugate optimization?
• What benchmark datasets exist for linker optimization or intracellular trafficking prediction?

Without such details, the AI-related sections remain conceptually broad and scientifically superficial.

 

2. Insufficient discussion of experimental validation and translational evidence

A major limitation is the lack of experimentally validated case studies demonstrating that AI-designed conjugates outperform conventionally engineered systems.

The manuscript repeatedly claims that AI:

• improves affinity,
• reduces toxicity,
• predicts resistance,
• and enhances intracellular delivery,

yet provides little direct evidence supporting these assertions in gynecological cancer applications.

The authors should discuss:

• actual AI-designed ADCs or peptide-drug conjugates that entered preclinical or clinical evaluation,
• quantitative improvements achieved by AI optimization,
• comparative validation versus conventional screening,
• and failure cases where AI predictions did not translate biologically.

Currently, many claims appear aspirational rather than evidence-based.

 

3. The discussion of intracellular trafficking and resistance prediction is overly speculative

Sections discussing intracellular trafficking, endosomal escape, resistance evolution, and adaptive redesign are among the most ambitious parts of the manuscript, yet they are also the least substantiated scientifically.

Predicting:

• endocytic pathway selection,
• lysosomal routing,
• payload release kinetics,
• and resistance emergence

remains an exceptionally difficult systems biology problem.

The manuscript should acknowledge:

• the current limitations of available datasets,
• the scarcity of high-quality longitudinal intracellular trafficking data,
• and the challenges in modeling dynamic tumor evolution.

At present, these sections sometimes overstate the practical maturity of current AI capabilities.

 

4. Lack of critical discussion regarding data quality and bias

Although the manuscript briefly mentions data limitations, the discussion is insufficient considering how central data quality is to AI performance.

The authors should more critically discuss:

• dataset imbalance,
• annotation inconsistency,
• experimental variability,
• overfitting risk,
• limited gynecological cancer datasets,
• and reproducibility challenges.

Additionally, many available structural and omics datasets are heavily biased toward well-studied antigens (e.g., HER2), potentially limiting generalizability to rare gynecological tumor subtypes.

This issue deserves substantially deeper treatment.

 

5. The manuscript insufficiently distinguishes between theoretical potential and clinically achievable implementation

Throughout the article, there is a tendency to blur the distinction between:

• what AI might theoretically enable,
• and what is currently feasible in translational oncology.

For example, patient-specific adaptive conjugate redesign based on real-time resistance prediction is presented almost as an emerging reality, whereas such workflows remain largely conceptual and are not clinically established.

The manuscript would benefit from clearer stratification of:

• current clinical reality,
• preclinical experimental capability,
• and long-term future vision.

Without this distinction, the review risks overstating the maturity of the field.

 

6. Regulatory and ethical discussions are underdeveloped

The section on regulatory considerations is relatively brief compared with the central role AI would play in therapeutic design.

Important issues requiring deeper discussion include:

• model explainability requirements,
• reproducibility standards,
• validation pathways for AI-assisted molecular design,
• intellectual property concerns,
• accountability in AI-generated therapeutic decisions,
• and regulatory expectations from agencies such as FDA or EMA.

Given the growing importance of AI governance in medicine, this section should be significantly expanded.

 

Minor Comments

Several statements throughout the manuscript are repetitive, particularly regarding:

• “AI transforming precision oncology,”
• “reducing off-target toxicity,”
• and “overcoming tumor heterogeneity.”

The manuscript would benefit from tighter editing to improve readability and reduce redundancy.

 

Some sections contain grammatical inconsistencies and awkward phrasing. Careful language editing by a fluent English scientific editor is recommended.

Examples include:

• “fall on structural, chemical, and biological data”
• “can manoeuvre through the complexity”
• “drug release the least”

These phrases reduce overall clarity.

 

Table 1 and Table 2 are useful summaries, but they remain relatively high-level. Including:

• representative AI platforms,
• specific software tools,
• public datasets,
• or experimentally validated examples

would greatly improve practical value.

 

The figures are visually attractive but conceptually simplified. The illustrations resemble graphical abstracts rather than scientifically detailed schematics.

For example:

• Figure 2 could include actual computational workflows,
• AI model architecture examples,
• or intracellular trafficking pathways with mechanistic detail.

 

The manuscript would benefit from a dedicated section discussing:

• multimodal foundation models,
• generative AI for biomolecule engineering,
• reinforcement learning approaches,
• and AI-guided de novo linker/payload generation.

These represent rapidly evolving areas currently underrepresented in the review.

Author Response

Reviewer # 3
This perspective addresses a timely and important topic at the intersection of artificial intelligence (AI), biomolecule–drug conjugates, and gynecological oncology. The manuscript provides a broad conceptual overview of how AI may facilitate affinity prediction, linker optimization, payload engineering, intracellular trafficking analysis, and resistance prediction in next-generation conjugate therapeutics. The topic is highly relevant given the increasing interest in AI-assisted precision oncology. However, despite its conceptual appeal, the manuscript currently remains largely descriptive and speculative. Many sections summarize general expectations of AI without providing sufficient technical depth, mechanistic insight, critical evaluation, or translational evidence. The article frequently uses broad statements regarding the transformative capability of AI, yet lacks detailed discussion of actual implemented workflows, validated case studies, model limitations, benchmarking strategies, and regulatory feasibility. As a result, the manuscript reads more as a generalized narrative overview rather than a rigorous perspective grounded in current technological realities.
Substantial revision is recommended before the manuscript can be considered for publication.

Response:    We sincerely thank the reviewer for the thorough and insightful evaluation of our manuscript.
We appreciate the reviewer’s recognition of the relevance and timeliness of the topic, particularly the growing importance of AI-assisted precision oncology in gynecological cancer therapeutics. We also acknowledge the concern that the original version of the manuscript was comparatively broad and, in certain sections, overly descriptive with limited mechanistic depth, translational discussion, and critical evaluation. In response, the manuscript has undergone substantial revision to improve its scientific rigor, technical depth, and translational relevance. Generalized and speculative statements regarding the transformative capabilities of AI have been carefully revised throughout the manuscript and replaced with more balanced, evidence-based descriptions. The revised version now provides expanded discussion of AI-assisted workflows involved in biomolecule-target affinity prediction, structural interaction modelling, linker and payload optimization, intracellular trafficking analysis, and resistance-associated therapeutic modelling.

To strengthen scientific balance, the manuscript now includes a more critical discussion of current limitations in AI-assisted conjugate engineering, including restricted gynecological cancer-specific datasets, model interpretability challenges, lack of standardized benchmarking frameworks, limited experimental and prospective clinical validation, reproducibility concerns, and translational feasibility constraints. Comparative discussion between conventional empirical approaches and AI-assisted design strategies has also been expanded to avoid overstating the current maturity of the field. In addition, the revised manuscript now incorporates more detailed gynecological cancer-specific discussion, including ovarian cancer heterogeneity, HPV-associated cervical cancer biomarkers, and molecular diversity in endometrial cancer, thereby improving disease-focused relevance and clinical context (Sec 2.6, 3.2). Collectively, these revisions substantially strengthen the manuscript by improving methodological depth, critical evaluation, translational grounding, and overall scientific balance, thereby aligning the review more closely with current technological realities and emerging clinical applicability in AI-guided biomolecule–drug conjugate engineering.

Major Comments
1.    The manuscript lacks technical depth regarding AI methodologies
The manuscript repeatedly refers to “AI,” “machine learning,” and “deep learning” as transformative tools, but the discussion remains highly generic throughout most sections. For example, CNNs, transformers, GNNs, and transfer learning are mentioned, yet the manuscript does not adequately explain:
•    what specific inputs are used,
•    what outputs are predicted,
•    how models are trained,
•    what datasets are required,
•    what performance metrics are used,
•    or how uncertainty is quantified.
The article would benefit substantially from concrete examples of implemented AI frameworks used in antibody-drug conjugate (ADC) or biomolecule–drug conjugate optimization. For instance:
•    Which GNN architectures are most suitable for affinity prediction?
•    How are structural embeddings generated?
•    Are AlphaFold-derived structures sufficiently accurate for conjugate optimization?
•    What benchmark datasets exist for linker optimization or intracellular trafficking prediction?
Without such details, the AI-related sections remain conceptually broad and scientifically superficial.
 
Response:    We sincerely thank the reviewer for this important and technically insightful comment. We acknowledge that the original version of the manuscript provided a comparatively broad conceptual overview of AI-assisted biomolecule–drug conjugate engineering with limited methodological detail regarding specific AI architectures, datasets, training strategies, and performance evaluation approaches. In response, the manuscript has been substantially revised to improve the technical depth, mechanistic clarity, and methodological specificity of the AI-related discussion. Specifically, Section 3.2 (“Affinity Prediction DL Architectures”) has been expanded to include additional discussion regarding the functional roles of convolutional neural networks (CNNs), graph neural networks (GNNs), transformer-based architectures, transfer learning approaches, and AlphaFold-assisted structural modelling in biomolecule–target affinity prediction and conjugate optimization. The revised text now clarifies representative model inputs (e.g., molecular graphs, protein sequences, docking conformations, structural embedding’s, and omics datasets), predicted outputs (e.g., binding affinity, linker stability, intracellular trafficking behavior, and resistance-associated therapeutic response), and commonly used evaluation parameters including prediction accuracy, area under the curve (AUC), loss minimization, and model generalizability.

In addition, Sections 3.3-3.6 have been revised to provide improved mechanistic discussion regarding affinity landscape modelling, dynamic binding kinetics, transfer learning under limited gynecological cancer datasets, and explainable AI approaches for biologically interpretable prediction systems. Sections 5 and 6 have also been strengthened through incorporation of more balanced discussion regarding intracellular trafficking prediction, linker–payload co-optimization, resistance-associated therapeutic modelling, and the current limitations of available datasets and experimental validation frameworks. Furthermore, a new summary table entitled “Representative AI methodologies in biomolecule–drug conjugate design for gynecological cancers” has been incorporated to provide a structured overview of major AI architectures, representative input data types, predicted outputs, and translational applications in conjugate engineering. The revised manuscript also now includes expanded discussion regarding current limitations associated with AI-assisted conjugate optimization, including restricted benchmark datasets, uncertainty quantification challenges, variability in structural prediction reliability, limited gynecological cancer-specific training cohorts, and the evolving translational reliability of computationally predicted therapeutic systems. Collectively, these revisions substantially strengthen the technical rigor, methodological depth, and translational relevance of the manuscript while maintaining the perspective-oriented scope of the review.

2.    Insufficient discussion of experimental validation and translational evidence
A major limitation is the lack of experimentally validated case studies demonstrating that AI-designed conjugates outperform conventionally engineered systems. The manuscript repeatedly claims that AI:
•    improves affinity,
•    reduces toxicity,
•    predicts resistance,
•    and enhances intracellular delivery,
yet provides little direct evidence supporting these assertions in gynecological cancer applications.
The authors should discuss:
•    actual AI-designed ADCs or peptide-drug conjugates that entered preclinical or clinical evaluation,
•    quantitative improvements achieved by AI optimization,
•    comparative validation versus conventional screening,
•    and failure cases where AI predictions did not translate biologically.
Currently, many claims appear aspirational rather than evidence-based.
 
Response:    We sincerely thank the reviewer for this important and insightful comment. We agree that experimental validation and translational evidence are essential for critically assessing the practical applicability of AI-assisted biomolecule–drug conjugate engineering. In response, the manuscript has been revised to provide a more balanced, evidence-based, and translationally grounded discussion regarding the current capabilities and limitations of AI-guided conjugate optimization in gynecological oncology.

Specifically, Section 7 (“Limitations and Challenges”) has been substantially expanded to emphasize that many currently available AI-assisted biomolecule–drug conjugate optimization frameworks remain predominantly in computational or early preclinical stages, with comparatively limited prospective biological and clinical validation in gynecological cancer settings. The revised manuscript now more clearly distinguishes experimentally validated findings from emerging computational predictions and explicitly acknowledges that discrepancies may arise between predicted molecular interactions and actual biological performance because of tumor heterogeneity, intracellular trafficking variability, microenvironmental complexity, and incomplete experimental datasets. In addition, Sections 5 and 6 have been revised to incorporate more critical discussion regarding the current translational limitations of AI-assisted linker optimization, intracellular trafficking prediction, payload release modelling, and resistance-associated therapeutic prediction systems. The revised text now emphasizes that many proposed AI-guided optimization strategies remain exploratory and require rigorous experimental benchmarking and prospective validation before reliable clinical implementation. Furthermore, generalized statements implying definitive superiority of AI-assisted approaches have been carefully revised throughout the manuscript and replaced with more cautious, evidence-based descriptions. Comparative discussion between conventional empirical approaches and AI-assisted conjugate engineering has also been strengthened in Section 7 and summarized in Table 3 to provide a more balanced assessment of current translational feasibility and benchmarking limitations. Collectively, these revisions strengthen the scientific rigor, translational realism, and critical evaluation presented throughout the manuscript.

3.    The discussion of intracellular trafficking and resistance prediction is overly speculative
Sections discussing intracellular trafficking, endosomal escape, resistance evolution, and adaptive redesign are among the most ambitious parts of the manuscript, yet they are also the least substantiated scientifically.
Predicting:
•    endocytic pathway selection,
•    lysosomal routing,
•    payload release kinetics,
•    and resistance emergence
remains an exceptionally difficult systems biology problem. The manuscript should acknowledge:
•    the current limitations of available datasets,
•    the scarcity of high-quality longitudinal intracellular trafficking data,
•    and the challenges in modeling dynamic tumor evolution.
At present, these sections sometimes overstate the practical maturity of current AI capabilities.

Response:    We sincerely thank the reviewer for this important and insightful comment. We agree that intracellular trafficking prediction, endosomal escape modelling, payload release kinetics, and resistance evolution remain highly complex and incompletely resolved systems biology challenges. In the original version of the manuscript, certain discussions within these sections may have conveyed a greater degree of technological maturity than is currently supported by available experimental and translational evidence. In response, the manuscript has been carefully revised to provide a more balanced, scientifically grounded, and translationally realistic discussion of these advanced AI-assisted applications. Specifically, Sections 6.1 (“Predicting Internalization Pathways and Endocytic Efficiency”), 6.2 (“Modelling Intracellular Trafficking and Payload Release”), and 6.3 (“Anticipating Resistance Mechanisms and Adaptive Tumor Responses”) have been substantially revised to acknowledge the limited availability of standardized longitudinal intracellular trafficking datasets, the scarcity of experimentally validated intracellular transport data, and the biological complexity associated with modelling dynamic tumor evolution and adaptive resistance mechanisms.

The revised manuscript now further emphasizes that intracellular routing behavior, lysosomal escape, payload release kinetics, and resistance-associated signaling changes are strongly influenced by tumor heterogeneity, temporal microenvironmental variability, intracellular signaling rewiring, and multifactorial adaptive responses that remain difficult to comprehensively capture using currently available computational frameworks. Additional discussion has also been incorporated regarding the exploratory and predominantly preclinical nature of many currently available AI-assisted intracellular trafficking and resistance prediction systems. Furthermore, generalized statements implying definitive predictive capability have been carefully revised throughout Sections 5 and 6 and replaced with more cautious, evidence-based descriptions emphasizing the emerging status and current translational limitations of these approaches. Collectively, these revisions improve the scientific balance, methodological accuracy, and translational realism of the manuscript.

4.    Lack of critical discussion regarding data quality and bias
Although the manuscript briefly mentions data limitations, the discussion is insufficient considering how central data quality is to AI performance. The authors should more critically discuss:
•    dataset imbalance,
•    annotation inconsistency,
•    experimental variability,
•    overfitting risk,
•    limited gynecological cancer datasets,
•    and reproducibility challenges.
Additionally, many available structural and omics datasets are heavily biased toward well-studied antigens (e.g., HER2), potentially limiting generalizability to rare gynecological tumor subtypes. This issue deserves substantially deeper treatment.

Response:    We sincerely thank the reviewer for this valuable and technically important comment. We fully agree that data quality, dataset bias, annotation inconsistency, and reproducibility limitations represent major challenges affecting the reliability, robustness, and translational applicability of AI-assisted biomolecule–drug conjugate engineering. In response, Section 2.6 has been substantially expanded to incorporate a more critical and detailed discussion regarding the limitations of currently available structural, omics, biochemical, and clinical datasets used for AI model development in gynecological oncology. The revised manuscript now specifically addresses challenges associated with dataset imbalance, experimental variability, incomplete annotation, overfitting risk, limited gynecological cancer-specific training cohorts, and reproducibility constraints across independent studies.

Additional emphasis has also been placed on the disproportionate representation of extensively studied antigens such as HER2, FRα, and EGFR within publicly available datasets, which may bias predictive performance and limit model generalizability toward rare gynecological tumor subtypes and under-characterized biomarkers. Furthermore, the revised text now discusses how variability in sequencing platforms, imaging methodologies, molecular interaction assays, and experimental reporting standards may introduce systematic bias and affect cross-study reproducibility and model reliability. To provide a more balanced and translationally grounded perspective, the revised manuscript also highlights the need for standardized multicenter datasets, harmonized annotation protocols, explainable AI frameworks, disease-specific gynecological cancer repositories, and rigorous external validation strategies to improve the reliability and clinical applicability of future AI-assisted conjugate engineering systems. Collectively, these revisions substantially strengthen the manuscript’s critical evaluation of current data-related limitations and improve the overall scientific rigor of the review.

5.    The manuscript insufficiently distinguishes between theoretical potential and clinically achievable implementation
Throughout the article, there is a tendency to blur the distinction between:
•    what AI might theoretically enable,
•    and what is currently feasible in translational oncology.
For example, patient-specific adaptive conjugate redesign based on real-time resistance prediction is presented almost as an emerging reality, whereas such workflows remain largely conceptual and are not clinically established. The manuscript would benefit from clearer stratification of:
•    current clinical reality,
•    preclinical experimental capability,
•    and long-term future vision.
Without this distinction, the review risks overstating the maturity of the field.

Response:    We sincerely thank the reviewer for this thoughtful and important comment. We fully agree that a clear distinction between theoretical AI potential, emerging preclinical capability, and clinically established implementation is essential to maintain scientific balance and avoid overstating the maturity of the field. In response, the manuscript has been carefully revised throughout to more explicitly differentiate between currently feasible translational applications and longer-term conceptual possibilities in AI-assisted biomolecule–drug conjugate engineering. Several statements implying near-clinical implementation of adaptive AI-guided therapeutic redesign have been moderated and replaced with more cautious, evidence-based descriptions reflecting the predominantly exploratory and preclinical status of many existing approaches.
Specifically, revisions have been incorporated in Section 6 (AI-Enabled Prediction of Cellular Internalization, Intracellular Trafficking, and Resistance Dynamics) and Section 8 (Future Perspectives: AI as a Catalyst for Precision Conjugate Oncology in Gynecological Cancers) to clarify that advanced concepts such as real-time resistance prediction, adaptive conjugate redesign, dynamically evolving therapeutic optimization, and patient-specific AI-guided treatment modification remain largely investigational and are not yet established in routine clinical oncology practice. Additional discussion has also been added regarding the substantial experimental, regulatory, computational, and prospective clinical validation challenges that must be addressed before such systems can become clinically feasible. Collectively, these revisions provide a more balanced, translationally grounded, and clinically realistic perspective on the current status and future trajectory of AI-guided biomolecule–drug conjugate development in gynecological oncology.

6.    Regulatory and ethical discussions are underdeveloped
The section on regulatory considerations is relatively brief compared with the central role AI would play in therapeutic design. Important issues requiring deeper discussion include:
•    model explainability requirements,
•    reproducibility standards,
•    validation pathways for AI-assisted molecular design,
•    intellectual property concerns,
•    accountability in AI-generated therapeutic decisions,
•    and regulatory expectations from agencies such as FDA or EMA.
Given the growing importance of AI governance in medicine, this section should be significantly expanded.

Response:    We sincerely thank the reviewer for this highly valuable and constructive comment. We fully agree that regulatory, ethical, and governance considerations are central to the responsible development and clinical translation of AI-assisted biomolecule–drug conjugate design and therefore warrant more comprehensive discussion. In response, the manuscript has been substantially revised to strengthen and expand the regulatory and ethical framework of AI-guided conjugate engineering. Specifically, Section 8.3 (Building a Collaborative and Responsible AI Ecosystem) has been expanded to provide a more structured and critical discussion of key translational governance dimensions, including model explainability requirements, reproducibility standards, validation pipelines for AI-assisted molecular design, intellectual property considerations, accountability frameworks for AI-generated therapeutic decisions, and evolving regulatory expectations from agencies such as the FDA and EMA. A clearer distinction has been introduced between computational innovation and regulatory-grade clinical translation, emphasizing that while AI-based conjugate design offers significant theoretical and preclinical promise, its clinical deployment requires rigorous validation, standardized benchmarking, transparent reporting, and adherence to evolving regulatory frameworks. Collectively, these revisions significantly strengthen the manuscript by providing a more balanced, transparent, and translationally grounded discussion of ethical and regulatory challenges in AI-guided gynecological oncology therapeutics.

Minor Comments
Several statements throughout the manuscript are repetitive, particularly regarding:
•    “AI transforming precision oncology,”
•    “reducing off-target toxicity,”
•    and “overcoming tumor heterogeneity.”
The manuscript would benefit from tighter editing to improve readability and reduce redundancy.
 
Some sections contain grammatical inconsistencies and awkward phrasing. Careful language editing by a fluent English scientific editor is recommended.
Examples include:
•    “fall on structural, chemical, and biological data”
•    “can manoeuvre through the complexity”
•    “drug release the least”
These phrases reduce overall clarity.
 
Table 1 and Table 2 are useful summaries, but they remain relatively high-level. Including:
•    representative AI platforms,
•    specific software tools,
•    public datasets,
•    or experimentally validated examples
would greatly improve practical value.
 
The figures are visually attractive but conceptually simplified. The illustrations resemble graphical abstracts rather than scientifically detailed schematics.
For example:
•    Figure 2 could include actual computational workflows,
•    AI model architecture examples,
•    or intracellular trafficking pathways with mechanistic details.
The manuscript would benefit from a dedicated section discussing:
•    multimodal foundation models,
•    generative AI for biomolecule engineering,
•    reinforcement learning approaches,
•    and AI-guided de novo linker/payload generation.
These represent rapidly evolving areas currently underrepresented in the review.

Response:    We sincerely thank the reviewer for these valuable suggestions aimed at improving the clarity, readability, and technical presentation of the manuscript. We fully agree that the original version contained repetitive phrasing, minor grammatical inconsistencies, and relatively schematic representations that required refinement. In response, the manuscript has undergone extensive language editing to improve readability and eliminate redundancy. In particular, repeated expressions such as “AI transforming precision oncology,” “reducing off-target toxicity,” and “overcoming tumor heterogeneity” have been systematically revised and redistributed across sections to ensure greater textual diversity and scientific precision without loss of meaning. Grammatical and stylistic corrections have also been implemented throughout the manuscript to improve clarity and scientific tone. Specific problematic phrases, including “fall on structural, chemical, and biological data,” “can manoeuvre through the complexity,” and “drug release the least,” have been revised to more accurate and grammatically appropriate formulations. 

Furthermore, Tables have been substantially strengthened by incorporating more application-oriented content, including representative AI methodologies, computational frameworks, and experimentally validated or widely used approaches where applicable. This enhancement improves their practical utility and translational relevance rather than remaining purely conceptual summaries. In addition, Figure 2 has been conceptually refined to better represent computational AI workflows, including explicit model architecture representation (CNNs, GNNs, transformers), data input modalities, and mechanistic biological processes such as receptor binding, internalization, trafficking, and payload release. This improves the scientific depth and mechanistic clarity of the illustration. Collectively, these revisions significantly improve the manuscript’s readability, technical accuracy, and conceptual depth while preserving its perspective-oriented structure.

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The revised version of the manuscript was significantly improved by the authors. The comments were addressed properly. I recommend acceptance of the manuscript for publication in the revised form.