Artificial Intelligence in Nanopharmaceutical Development: From Predictive Design to Clinical Translation

Round 1

Reviewer 1 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

The revised manuscript is significantly improved: it now includes quantitative metrics, systematic method comparisons, concrete data standardization proposals, a tiered validation framework, practical uncertainty quantification techniques, a realistic outlook on precision nanomedicine, and clear prioritization of mature vs speculative AI applications.

Reviewer 2 Report (Previous Reviewer 5)

Comments and Suggestions for Authors

The manuscript by Gonçalves has been significantly improved during the revision. The author has addressed the referees' comments, and the paper is now suitable for publication, albeit a smooth grammatical revision is suggested (lines 67, 138, last column of the last item /Table 2/ on page 10, 490, 699, 791, 1134, 1253, 1366, 1367, 1370, 1411, 1509, 1677).

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This review provides a comprehensive overview of current AI-based methods and discusses how AI may support nanopharmaceutical development, from predictive formulation design to clinical translation. The manuscript is generally well organized and covers a broad range of relevant topics. However, several issues should be addressed before publication:

1. It would strengthen the manuscript if the authors can include representative tools, platforms, or models associated with each AI-based approach discussed in the review.
2. Although the manuscript discusses multiple application areas of AI in nanopharmaceutical research (e.g., nanocarrier design, formulation optimization, pharmacokinetics, and toxicity prediction), these sections are currently presented in a relatively conceptual manner. It would strengthen the review if the authors could provide specific examples from the literature demonstrating how AI-based approaches have been practically applied in these areas, including whether experimental validation was performed and what outcomes or improvements were achieved.
3. Some examples appear repetitive across sections. For example, the discussions involving PLGA and lipid nanoparticles in lines 159-162 and 242-246 are highly similar. The authors may consider consolidating or differentiating these examples to improve readability.
4. The title of Section 2, “Advanced Modeling Approaches for Nanopharmaceutical Design,” may not fully reflect the content of Sections 2.10-2.15, which are more closely related to nano-bio interactions and pharmacokinetics. The authors may consider either revising the section title or reorganizing these subsections into a separate section for better structural clarity.
5. Abbreviations should be defined upon first appearance in the manuscript. For example, “DoE” in lines 255 (but not line 260) should be written in full initially. The authors are encouraged to carefully check the manuscript throughout for similar issues.

Author Response

Response to Reviewer 1

 

1. It would strengthen the manuscript if the authors can include representative tools, platforms, or models associated with each AI-based approach discussed in the review.

Response: We thank the reviewer for this valuable suggestion. We agree that the inclusion of representative tools, platforms, and model architectures would strengthen the practical value of the review and make the discussion more useful for readers interested in implementing AI-based strategies in nanopharmaceutical development. Accordingly, we have revised the manuscript by adding representative examples of tools, platforms, and models associated with the main AI-based approaches discussed, including classical machine learning, deep learning, graph-based learning, Bayesian and evolutionary optimization, physics-informed and hybrid modeling, digital twins, process analytical technology, explainable AI, and pharmacokinetic/toxicity prediction models. In addition, we included a new summary table linking each AI-based approach to typical input data, predicted outputs, representative tools or platforms, and relevant nanopharmaceutical applications. This addition improves the implementation-oriented perspective of the review and helps bridge the conceptual discussion with practical applications. 

 

2. Although the manuscript discusses multiple application areas of AI in nanopharmaceutical research (e.g., nanocarrier design, formulation optimization, pharmacokinetics, and toxicity prediction), these sections are currently presented in a relatively conceptual manner. It would strengthen the review if the authors could provide specific examples from the literature demonstrating how AI-based approaches have been practically applied in these areas, including whether experimental validation was performed and what outcomes or improvements were achieved.

Response: We thank the reviewer for this important suggestion. We agree that the previous version discussed several AI applications in a relatively conceptual manner and that the review would benefit from more concrete examples demonstrating practical implementation. Accordingly, we revised the manuscript to include selected examples from the literature showing how AI-based approaches have been applied to nanopharmaceutical and nanomaterial research, including formulation–property prediction, machine learning-guided high-throughput nanoparticle optimization, protein corona-based biological fate prediction, pharmacokinetic and biodistribution modeling, and safety- or nanoactivity-related prediction. We also added a new summary table describing the AI approach used, the validation strategy, and the main outcome or improvement reported in each study. In addition, we incorporated specific examples throughout the relevant sections of the manuscript to clarify whether the approaches were based on retrospective validation, experimental feedback loops, protein corona-based biological validation, or comparison with measured pharmacokinetic profiles. These revisions make the discussion more practical and help readers better understand the current level of maturity, validation, and translational relevance of AI-based methods in nanopharmaceutical development. 

 

3. Some examples appear repetitive across sections. For example, the discussions involving PLGA and lipid nanoparticles in lines 159-162 and 242-246 are highly similar. The authors may consider consolidating or differentiating these examples to improve readability.

 

Response: We thank the reviewer for this helpful observation. We agree that some examples were repeated across sections, particularly those involving PLGA and lipid nanoparticles. To improve readability and avoid redundancy, we revised the manuscript by consolidating these examples. Specifically, the broader discussion in Section 2 was shortened and made more general, while the detailed discussion of PLGA and lipid nanoparticle examples was retained in Section 2.1, where formulation–property relationships and machine learning applications are discussed in greater detail. This revision improves the distinction between the general methodological overview and the specific application-focused discussion. 

 

4. The title of Section 2, “Advanced Modeling Approaches for Nanopharmaceutical Design,” may not fully reflect the content of Sections 2.10-2.15, which are more closely related to nano-bio interactions and pharmacokinetics. The authors may consider either revising the section title or reorganizing these subsections into a separate section for better structural clarity.

Response: We thank the reviewer for this helpful suggestion. We agree that the title of Section 2 did not fully reflect the scope of the subsections discussing nano–bio interactions, pharmacokinetics, toxicity, immunogenicity, and precision nanomedicine. To improve structural clarity, we reorganized the manuscript by retaining the modeling, optimization, PAT, and QbD-related content within Section 2 and moving the nano–bio interaction and pharmacokinetic content into a new independent Section 3, entitled “AI in Nano–Bio Interactions, Pharmacokinetics, and Precision Nanomedicine.” The former subsections 2.10–2.15 were renumbered as Sections 3.1–3.6, and the subsequent sections were renumbered accordingly. This reorganization better distinguishes AI applications in formulation and process development from those related to biological behavior, safety, pharmacokinetics, and precision nanomedicine. 

 

5. Abbreviations should be defined upon first appearance in the manuscript. For example, “DoE” in lines 255 (but not line 260) should be written in full initially. The authors are encouraged to carefully check the manuscript throughout for similar issues.

Response: We thank the reviewer for this careful observation. We agree that all abbreviations should be defined upon first appearance to improve clarity and readability. Accordingly, we carefully checked the manuscript and revised the first occurrence of abbreviations throughout the text, tables, and figure captions. In particular, “DoE” was defined as “design of experiments (DoE)” at its first appearance, and subsequent mentions were kept in abbreviated form. We also checked and corrected other abbreviations, including AI/ML, PDI, PK, PBPK, QbD, PAT, CMAs, CPPs, CQAs, t-SNE, UMAP, SHAP, LIME, PLGA, PLGA-PEG, RSM, and NSGA-II. These revisions ensure that abbreviations are introduced consistently before being used throughout the manuscript. 

Reviewer 2 Report

Comments and Suggestions for Authors

Review of Draft: "Artificial Intelligence in Nanopharmaceutical Development: From Predictive Design to Clinical Translation" by Sonchini Renato

The article reviews how artificial intelligence (AI) is reshaping nanopharmaceutical development. It combines technical detail with practical relevance, connecting new algorithms with real-world, regulatory, and clinical issues. The manuscript covers many topics—such as machine learning, deep learning, hybrid and physics-informed models, digital twins, and process optimization—and provides a fair view of both the possibilities and drawbacks.

The article reviews the entire nanopharmaceutical process, from developing new formulations and improving processes to predicting biological effects, assessing toxicity, understanding how drugs act in the body, and complying with regulations. Each modeling method is explained, with its benefits, limits, and uses.

The discussion draws on current research and demonstrates deep knowledge of both computing and the pharmaceutical industry. Complex ideas such as hybrid models, digital twins, and multiscale modeling are clearly explained.

The review avoids exaggeration and honestly discusses current problems, including differences in data types, a lack of shared standards, difficulty understanding AI decisions, and regulatory issues.

Real-world translation, regulatory alignment, and clinical adoption are recurring themes, making the article valuable to researchers, developers, and regulators.

The manuscript flows logically, with clear sectioning and transitions, effective use of summary tables and figures (where present), and a strong narrative from fundamentals to the future outlook.

Areas for Improvement

Some sections are dense and extremely technical. Deliberate use of summary boxes, glossaries, schematic figures and brief case studies would improve accessibility, especially for interdisciplinar readers.

The authors explained the methods and challenges well, but adding a few concrete case studies or clear efficiency metrics that show real-world AI impact would make this section more useful.

Summarize the main best practices and biggest challenges at the end of the section so readers walk away with clear, memorable takeaways.

Building on the ethical, legal and social implications—data privacy, bias, explainability and patient consent—would enhance completeness, especially as AI moves towards patient-facing applications.

This article is a timely, authoritative and balanced resource on the rapidly developing intersection of AI and nanomedicine. With targeted refinements in accessibility, practical illustration and actionable recommendations, it will function as a valuable reference intended for advancing scholarship and practice in this field.

Specific comments

Introduction

The introduction sets the stage thoroughly for the review by outlining the challenges of nanopharmaceutical development and how AI is able to address them.

The translation gap between preclinical research and clinical implementation is well articulated, underscoring the need for forecasting and integrative computational approaches.

Clear explanation of how AI—including machine learning (ML: algorithms that enable computers to learn from data), deep learning (DL: multi-layered neural networks for complex data analysis), and hybrid modeling (integration of mechanistic and AI approaches)—can enhance traditional pharmaceutical development frameworks such as design of experiments (DoE), quality by design (QbD), and process analytical technology (PAT: real-time monitoring during manufacturing).

Appropriate references are cited to support key points, and the section situates the discussion within current regulatory and technological trends.

The progression from traditional challenges to current solutions to the promise of AI is logical and easy to follow.

Areas for Improvement

The introduction is rich with information, which clearly shows the author’s expertise. However, it might feel a bit dense for readers. Adding a few more subheadings or brief summary statements—especially at key paragraph breaks—could make it easier to follow and help guide readers through the main points.

Key Points Summary: End the introduction with a concise summary box that highlights the main messages before continuing.

Several phrases about current approach limitations (such as data fragmentation and lack of standardization) are repeated throughout. Consolidate these ideas for brevity and clarity.

Some terms, such as "hybrid mechanistic-AI frameworks," "digital twins" and "process analytical technology" appear before being defined. Give a quick definition in brackets or include them in a glossary or a footnote.

The introduction could be strengthened by including one or two brief, concrete examples of successful AI applications in nanopharmaceuticals (even as a teaser for later sections).

Minor Issues:

Some sentences are very long and complex. Try splitting them into shorter, clearer parts to make them easier to read.

While you mention regulatory and translational challenges, consider adding a sentence or two to the introduction that notes the ethical, privacy, and societal implications will be discussed later.

Sections 2–2.15

The sections are well organized, comprehensive, and up to date.

They are balanced, discerning tone and clear articulation of both advancements and barriers.

Points for Improvement:

Section summaries: Use summary boxes at the end of complex sections to clarify main concepts and boost accessibility.

Use more practical case studies and efficient measures to ground the discussion.

Section 3: Translational Challenges and regulatory perspectives.

The section appropriately places AI-driven nanopharmaceutical development within the wider context of clinical translation, manufacturing and regulatory systems.

It acknowledges that high model effectiveness alone is insufficient for translation, highlighting the need for reproducibility, scalability, regulatory transparency and real-life evidence of benefit.

Figure 5 clearly outlines the roadmap from the data generation to clinical implementation, helping readers visualize the translational journey and its key requirements.

The section systematically addresses critical challenges like data quality and heterogeneity, limited dataset size, descriptor standardization, lack of external validation, interpretability, uncertainty measurement, life cycle management, scalability and regulatory conformance are all examined.

It provides a detailed discussion of regulatory expectations for AI models, including documentation, validation, version control, and lifecycle management—elements crucial to the adoption of AI in pharmaceutical development.

Summary Translational and Oversight Challenges: Include an efficient summary box to reinforce impacts and recommended strategies.

Areas for Improvement

Although the challenges are well described, consider including brief real-world examples and mini-case studies that show how a specific barrier (e.g., data heterogeneity patterns or lack of external validation) affected an AI enabled nanomedicine project. Suggest more practical suggestions for overcoming each challenge, possibly as a bulleted list or highlighted box “Best Practices’’.

The section may benefit from a more granular overview of current regulatory pilot programs (e.g., the FDA’s Digital Health Software Precertification Program and EMA’s AI reflection paper) and their relevance to nanopharmaceuticals. Briefly discuss the changing policy landscape for self learning, continuously learning AI models in regulated environments.

Ethical and data privacy concerns are only briefly mentioned; consider developing informed consent, patient data governance and bias prevention—topics particularly relevant to precision nanomedicine. Point out the need for interdisciplinary collaboration among scientists, clinicians, regulators and philosophers.

Key Takeaways: Add a concise summary box ending the section on critical requirements for a successful translation.

Section 4: Future perspectives.

The section effectively anticipates the next frontiers of AI-powered nanopharmaceuticals, including closed-loop optimization, digital twins and precise nanomedicine.

There is a clear call for validated, interpretable and clinically relevant AI frameworks that move beyond proof-of-concept studies.

The text recognizes the need to integrate data from formulation, process, biological, manufacturing, and clinical domains and stresses the importance of multimodal data fusion for future progress.

It acknowledges real-world obstacles—such as model opacity, bias, overfitting, lack of standardization, and regulatory requirements—offering a balanced, pragmatic outlook.

The need for collaboration among academia, industry, clinicians, and regulators is clearly articulated; the section highlights requirements for standardized datasets, benchmarking, and validation protocols.

The review presents both the opportunities and hurdles of AI-driven, patient designed nanomedicine, thoughtfully addressing privacy, ethical concerns, and clinical responsibility.

Areas for Improvement

Please add a summary box that highlights the main priorities and that can provide a practical roadmap for the coming decade.

Include a specific example or a visionary case study (even a hypothetical one) that illustrates closed-loop development or digital twins in practice.

If possible, include a schematic overview figure or workflow diagram that visually maps out the “future pipeline” for AI-powered nanomedicine.

Briefly expand on the intersection of regulatory, clinical, and technical domains—what new skill sets or organizational structures will be required.

Briefly discuss the need for open, interoperable data standards and model repositories (e.g., MoleculeNet for molecules), as well as ongoing model monitoring, updating, and revalidation.

5: Conclusions

This section explains that the review’s main point is how AI is changing nanopharmaceutical development from slow trial-and-error work into a faster, more predictive, data-driven process that focuses on real-world applications.

It revisits all the major technical strategies discussed (machine learning, deep learning, physics-informed modeling, hybrid approaches, automated optimization, integration with DoE, QbD, PAT, and digital twins) and ties them to their impact on formulation, process, and biological prediction.

The conclusion emphasizes how AI really shines when it comes to modeling nano–bio interactions—things like protein corona formation, cellular uptake, pharmacokinetics, immunogenicity, and toxicity—where traditional methods often fall short.

The section acknowledges ongoing limitations—like fragmented data, inconsistent standards, limited validation, and challenges with interpretability and oversight—while emphasizing that getting the computation right is not enough on its own to ensure real-world impact.

In the end, the paper says that AI should not only be used as a computer tool but also as a means to help move research into real-world practice. It also says that AI work needs to be repeatable, clear, well-tested, and in line with current rules to have a real impact.

Author Response

Response to Reviewer 2

1. Introduction

The introduction sets the stage thoroughly for the review by outlining the challenges of nanopharmaceutical development and how AI is able to address them. The translation gap between preclinical research and clinical implementation is well articulated, underscoring the need for forecasting and integrative computational approaches. Clear explanation of how AI—including machine learning (ML: algorithms that enable computers to learn from data), deep learning (DL: multi-layered neural networks for complex data analysis), and hybrid modeling (integration of mechanistic and AI approaches)—can enhance traditional pharmaceutical development frameworks such as design of experiments (DoE), quality by design (QbD), and process analytical technology (PAT: real-time monitoring during manufacturing). Appropriate references are cited to support key points, and the section situates the discussion within current regulatory and technological trends. The progression from traditional challenges to current solutions to the promise of AI is logical and easy to follow.

Response: We thank the reviewer for this positive assessment of the Introduction. We are pleased that the reviewer found the section logically structured and effective in presenting the challenges of nanopharmaceutical development, the translational gap between preclinical and clinical implementation, and the potential role of AI-based approaches in addressing these issues. We also appreciate the recognition that the Introduction appropriately situates the review within current technological and regulatory trends. In response to the reviewer’s broader comments on abbreviation use, we carefully checked the manuscript to ensure that abbreviations such as AI, ML, DL, DoE, QbD, and PAT are defined upon first appearance. No major changes were required to the conceptual structure of the Introduction. 

 

2. Areas for Improvement

The introduction is rich with information, which clearly shows the author’s expertise. However, it might feel a bit dense for readers. Adding a few more subheadings or brief summary statements—especially at key paragraph breaks—could make it easier to follow and help guide readers through the main points. Key Points Summary: End the introduction with a concise summary box that highlights the main messages before continuing. Several phrases about current approach limitations (such as data fragmentation and lack of standardization) are repeated throughout. Consolidate these ideas for brevity and clarity. Some terms, such as "hybrid mechanistic-AI frameworks," "digital twins" and "process analytical technology" appear before  being defined. Give a quick definition in brackets or include them in a glossary or a footnote. The introduction could be strengthened by including one or two brief, concrete examples of successful AI applications in nanopharmaceuticals (even as a teaser for later sections).

Response: We thank the reviewer for these constructive suggestions. We agree that, although the Introduction provides a comprehensive overview of the field, its readability can be improved by adding clearer transition points, reducing repeated statements, defining technical terms upon first appearance, and including brief concrete examples of AI applications. Accordingly, we revised the Introduction to improve flow and accessibility. Specifically, we added short guiding statements to clarify the progression from translational challenges to AI-enabled solutions, reduced repeated wording related to data fragmentation and lack of standardization, and defined key terms such as hybrid mechanistic–AI frameworks, digital twins, and process analytical technology upon first appearance. We also added brief examples of practical AI applications in nanopharmaceutical research, including machine learning-based prediction of PLGA nanoparticle attributes and AI-assisted pharmacokinetic modeling for nanoparticle delivery. Finally, we added a concise key-points summary at the end of the Introduction to guide readers before the more detailed sections of the review. 

 

3. Minor Issues:

Some sentences are very long and complex. Try splitting them into shorter, clearer parts to make them easier to read. While you mention regulatory and translational challenges, consider adding a sentence or two to the introduction that notes the ethical, privacy, and societal implications will be discussed later.

Response: We thank the reviewer for this helpful suggestion. We agree that some sentences in the Introduction were relatively long and could be improved for readability. Accordingly, we revised selected sentences to make them shorter and clearer, while preserving the scientific content and logical flow of the section. We also added a brief statement in the Introduction indicating that ethical, privacy, and societal implications of AI-supported nanopharmaceutical development are considered later in the review. This addition helps prepare readers for the broader translational and governance-related discussion presented in the subsequent sections. 

 

4. Sections 2–2.15

The sections are well organized, comprehensive, and up to date. They are balanced, discerning tone and clear articulation of both advancements and barriers.

Response: We thank the reviewer for this positive assessment. We are pleased that the reviewer found these sections well organized, comprehensive, up to date, and balanced in tone. We also appreciate the recognition that the manuscript clearly articulates both the advances and barriers associated with AI-driven nanopharmaceutical development. In response to other reviewer comments, we further improved the structural clarity of these sections by separating the formulation/modeling content from the nano–bio interaction, pharmacokinetic, toxicity, and precision nanomedicine content. No major conceptual changes were required in response to this comment. 

 

5. Points for Improvement:

Section summaries: Use summary boxes at the end of complex sections to clarify main concepts and boost accessibility. Use more practical case studies and efficient measures to ground the discussion.

Response: We thank the reviewer for this helpful suggestion. We agree that brief section summaries can improve accessibility, especially in complex sections that discuss multiple AI approaches, validation requirements, and translational barriers. Accordingly, we added concise summary statements at the end of the major sections to highlight the main concepts and guide readers through the review. In addition, we strengthened the practical orientation of the manuscript by adding representative literature examples and efficiency-related outcomes in Table 2, including formulation–property prediction, high-throughput nanoparticle optimization, protein corona-based biological fate prediction, pharmacokinetic modeling, and safety- or nanoactivity-related prediction. These revisions help ground the conceptual discussion in practical applications and clarify the potential value of AI-based approaches for improving prediction, optimization, experimental efficiency, and translational decision-making. 

6. Section 3: Translational Challenges and regulatory perspectives.

The section appropriately places AI-driven nanopharmaceutical development within the wider context of clinical translation, manufacturing and regulatory systems. It acknowledges that high model effectiveness alone is insufficient for translation, highlighting the need for reproducibility, scalability, regulatory transparency and real-life evidence of benefit. Figure 5 clearly outlines the roadmap from the data generation to clinical implementation, helping readers visualize the translational journey and its key requirements. The section systematically addresses critical challenges like data quality and heterogeneity, limited dataset size, descriptor standardization, lack of external validation, interpretability, uncertainty measurement, life cycle management, scalability and regulatory conformance are all examined. It provides a detailed discussion of regulatory expectations for AI models, including documentation, validation, version control, and lifecycle management—elements crucial to the adoption of AI in pharmaceutical development. Summary Translational and Oversight Challenges: Include an efficient summary box to reinforce impacts and recommended strategies.

Response: We thank the reviewer for this positive and constructive assessment. We are pleased that the reviewer found the section effective in placing AI-driven nanopharmaceutical development within the broader context of clinical translation, manufacturing, and regulatory systems. We also appreciate the recognition that the section addresses key challenges such as data quality, dataset size, descriptor standardization, external validation, interpretability, uncertainty quantification, lifecycle management, scalability, and regulatory conformance. To further improve accessibility and reinforce the main translational and oversight messages, we revised the section by introducing Table 5 more explicitly as a concise summary of the main challenges, their impacts, recommended strategies, and translational relevance. We also revised the table caption to emphasize its function as a summary of translational and oversight challenges. 

7. Areas for Improvement

Although the challenges are well described, consider including brief real-world examples and mini-case studies that show how a specific barrier (e.g., data heterogeneity patterns or lack of external validation) affected an AI enabled nanomedicine project. Suggest more practical suggestions for overcoming each challenge, possibly as a bulleted list or highlighted box “Best Practices’’. The section may benefit from a more granular overview of current regulatory pilot programs (e.g., the FDA’s Digital Health Software Precertification Program and EMA’s AI reflection paper) and their relevance to nanopharmaceuticals. Briefly discuss the changing policy landscape for self learning, continuously learning AI models in regulated environments. Ethical and data privacy concerns are only briefly mentioned; consider developing informed consent, patient data governance and bias prevention—topics particularly relevant to precision nanomedicine. Point out the need for interdisciplinary collaboration among scientists, clinicians, regulators and philosophers. Key Takeaways: Add a concise summary box ending the section on critical requirements for a successful translation.

Response: We thank the reviewer for these valuable suggestions. We agree that the section can be strengthened by providing more practical guidance, clarifying the evolving regulatory landscape, and expanding the discussion of ethical and data governance issues. Accordingly, we revised the section to better connect translational barriers with practical mitigation strategies, including standardized reporting, external validation, uncertainty quantification, lifecycle monitoring, and interdisciplinary oversight. We also expanded the regulatory discussion to mention relevant policy developments, including the FDA’s completed Digital Health Software Precertification Pilot Program, the emerging role of predetermined change control plans for AI-enabled software, and the EMA reflection paper on the use of AI in the medicinal product lifecycle. In addition, we expanded the ethical and governance discussion to address informed consent, patient data governance, bias prevention, privacy protection, and the need for interdisciplinary collaboration among formulation scientists, data scientists, clinicians, regulators, ethicists, and other stakeholders. Finally, we revised Table 5 and the concluding statements of the section to function as a concise summary of best practices and key requirements for successful translation. 

8. Section 4: Future perspectives.

The section effectively anticipates the next frontiers of AI-powered nanopharmaceuticals, including closed-loop optimization, digital twins and precise nanomedicine. There is a clear call for validated, interpretable and clinically relevant AI frameworks that move beyond proof-of-concept studies. The text recognizes the need to integrate data from formulation, process, biological, manufacturing, and clinical domains and stresses the importance of multimodal data fusion for future progress. It acknowledges real-world obstacles—such as model opacity, bias, overfitting, lack of standardization, and regulatory requirements—offering a balanced, pragmatic outlook. The need for collaboration among academia, industry, clinicians, and regulators is clearly articulated; the section highlights requirements for standardizeddatasets, benchmarking, and validation protocols. The review presents both the opportunities and hurdles of AI-driven, patient designed nanomedicine, thoughtfully addressing privacy, ethical concerns, and clinical responsibility.

Areas for Improvement

Please add a summary box that highlights the main priorities and that can provide a practical roadmap for the coming decade. Include a specific example or a visionary case study (even a hypothetical one) that illustrates closed-loop development or digital twins in practice. If possible, include a schematic overview figure or workflow diagram that visually maps out the “future pipeline” for AI-powered nanomedicine. Briefly expand on the intersection of regulatory, clinical, and technical domains—what new skill sets or organizational structures will be required. Briefly discuss the need for open, interoperable data standards and model repositories (e.g., MoleculeNet for molecules), as well as ongoing model monitoring, updating, and revalidation.

Response: We thank the reviewer for this constructive suggestion. We agree that the Future Perspectives section would benefit from a clearer practical roadmap for the coming decade. Accordingly, we added a concise roadmap table summarizing the main priorities for future AI-driven nanopharmaceutical development, including open and interoperable datasets, model repositories, closed-loop experimentation, hybrid mechanistic–AI modeling, digital twins, regulatory-ready workflows, ethical governance, and interdisciplinary organizational structures. We also included a visionary practical scenario illustrating how closed-loop formulation development and digital twins could operate in an integrated nanopharmaceutical development pipeline. Rather than adding an additional schematic figure, we incorporated the future pipeline in a structured roadmap table to avoid visual redundancy with the existing figures and to keep the section concise and accessible. We also expanded the discussion of open data standards, model repositories, model monitoring, updating, revalidation, and the new interdisciplinary skill sets required at the interface of technical, clinical, regulatory, and ethical domains. 

9. Conclusions

This section explains that the review’s main point is how AI is changing nanopharmaceutical development from slow trial-and-error work into a faster, more predictive, data-driven process that focuses on real-world applications. It revisits all the major technical strategies discussed (machine learning, deep learning, physics-informed modeling, hybrid approaches, automated optimization, integration with DoE, QbD, PAT, and digital twins) and ties them to their impact on formulation, process, and biological prediction. The conclusion emphasizes how AI really shines when it comes to modeling nano–bio interactions—things like protein corona formation, cellular uptake, pharmacokinetics, immunogenicity, and toxicity—where traditional methods often fall short. The section acknowledges ongoing limitations—like fragmented data, inconsistent standards, limited validation, and challenges with interpretability and oversight—while emphasizing that getting the computation right is not enough on its own to ensure real-world impact. In the end, the paper says that AI should not only be used as a computer tool but also as a means to help move research into real-world practice. It also says that AI work needs to

Response: We thank the reviewer for this positive assessment of the Conclusions section. We are pleased that the reviewer found that the section effectively summarizes the central message of the review, namely that AI can help shift nanopharmaceutical development from empirical trial-and-error approaches toward more predictive, data-driven, and translationally oriented strategies. We also appreciate the recognition that the Conclusions appropriately integrate the major technical themes discussed throughout the manuscript, including machine learning, deep learning, physics-informed modeling, hybrid approaches, automated optimization, DoE, QbD, PAT, digital twins, nano–bio interaction modeling, pharmacokinetics, toxicity, and clinical translation. No major changes were required in response to this comment. 

 

Reviewer 3 Report

Comments and Suggestions for Authors

refer attachment

Comments for author File: Comments.pdf

Author Response

Response to Reviewer 3

Overview This review examined how AI encompassing ML, DL, physics-informed modeling, and hybrid approaches, can support nanopharmaceutical development from predictive formulation design and multi-objective optimization to modeling of nano-bio interactions, pharmacokinetics, and toxicity. However, successful clinical translation remains limited by fragmented datasets, lack of external validation, poor model interpretability, and evolving regulatory expectations, requiring standardized data infrastructures and validated, explainable AI frameworks. 

Comments/Suggestions 

1. The manuscript comprehensively covered AI applications across the entire nanopharmaceutical pipeline, from design to clinical translation. The length leads to superficial treatment of many topics; the paper lacks a clear hierarchy of which AI strategies are most mature or impactful. Author should explicitly prioritize areas where AI has already shown validated preclinical or translational value vs speculative applications. 

Response: We thank the reviewer for this important observation. We agree that, because the review covers AI applications across the entire nanopharmaceutical pipeline, it is important to distinguish between approaches that already have stronger experimental, preclinical, or translational support and those that remain more exploratory or future-oriented. Accordingly, we revised the manuscript to make this hierarchy more explicit. We now emphasize that AI applications such as formulation-property prediction, high-throughput formulation optimization, protein corona-based biological fate prediction, and AI-assisted pharmacokinetic modeling currently represent more mature or experimentally supported areas. In contrast, applications such as fully autonomous closed-loop development, patient-specific digital twins, reinforcement learning-based process control, and precision nanomedicine decision-support systems remain promising but less clinically validated. This prioritization helps readers distinguish validated and near-term opportunities from more speculative future directions. 

 

2. The review identified multiple AI paradigms (supervised learning, DL, physics-informed modeling, Bayesian optimization, etc.). However, it does not systematically compare their relative performance, data requirements, or suitability for specific nanopharmaceutical problems. The author should add a comparative analysis table to guide readers on selecting the right AI method for a given formulation or biological prediction task. 

 

Response: We thank the reviewer for this important suggestion. We agree that, although the manuscript discusses multiple AI paradigms, readers would benefit from clearer guidance on their relative maturity, data requirements, interpretability, and suitability for different nanopharmaceutical tasks. Because the revised manuscript already contains several summary tables, we avoided adding another table and instead incorporated a concise comparative decision-guidance paragraph in Section 2. This new text explicitly distinguishes supervised machine learning, deep learning, Bayesian and evolutionary optimization, physics-informed and hybrid mechanistic–AI models, reinforcement learning, digital twins, and closed-loop systems according to their typical data requirements, practical maturity, interpretability, and most appropriate applications. This revision helps readers identify which AI strategies are currently more suitable for formulation-property prediction, image and multimodal data analysis, optimization, pharmacokinetic modeling, manufacturing support, and future autonomous or precision nanomedicine applications. 

 

3. The manuscript cited many representative studies and discusses predictive accuracy. It rarely reports quantitative performance metrics (e.g., R2 , RMSE, sensitivity/specificity) across studies, making it impossible to assess how well AI actually performs. Adding a summary of reported prediction errors and validation outcomes would strengthen the article. 

 

Response: We thank the reviewer for this important suggestion. We agree that reporting quantitative performance metrics and validation outcomes helps readers evaluate the actual predictive value of AI-based approaches. Accordingly, we revised the manuscript to clarify that performance metrics are not uniformly reported across the representative studies, which itself reflects an important limitation in the field. Where available, we added quantitative examples of model performance and validation outcomes, including R² and RMSE values for AI-assisted PBPK modeling and comparative findings showing improved RMSE and coefficient of determination for machine learning models relative to design-of-experiments models in PLGA nanoparticle optimization. We also clarified when studies relied mainly on retrospective validation, prediction graphs, experimental feedback, or comparison with measured pharmacokinetic profiles. These revisions provide a more balanced assessment of AI performance and highlight the need for standardized reporting of predictive metrics, external validation, and uncertainty estimates in future studies. 

 

4. The review stated that many AI models predict particle size, zeta potential, and encapsulation efficiency. It does not give sufficient attention to how often these physicochemical predictions fail to translate into meaningful biological or clinical outcomes. The manuscript should explicitly quantify the gap between AI-predicted CQAs and actual in vivo performance, and call for more models that integrate biological endpoints from the start. 

 

Response: We thank the reviewer for this important comment. We agree that prediction of physicochemical critical quality attributes, such as particle size, zeta potential, polydispersity index, and encapsulation efficiency, should not be interpreted as direct evidence of biological or clinical performance. To address this issue, we revised the manuscript to explicitly discuss the gap between CQA prediction and in vivo outcomes. We now emphasize that CQA prediction is primarily useful for early formulation screening and process understanding, but that meaningful translation requires integration with biological endpoints. We also added a quantitative example illustrating this translational gap: despite rational physicochemical optimization, previous analyses of nanoparticle delivery to solid tumors have shown that only a small fraction of the injected dose reaches the target site, with reported median tumor delivery efficiencies of approximately 0.7% of the injected dose. This example highlights that formulation-level optimization alone is insufficient to ensure therapeutic performance. Accordingly, we revised the manuscript to call for AI models that integrate biological endpoints from the beginning, including protein corona formation, cellular uptake, biodistribution, pharmacokinetics, immune activation, toxicity, and therapeutic response. We also clarified the regulatory implications of this issue by noting that AI models should define their intended use and distinguish whether they support early formulation screening, process control, biological prediction, or clinical decision-making.

5. The manuscript acknowledged that fragmented and heterogeneous datasets limit model generalizability. However, it does not provide concrete, actionable recommendations for data standardization (e.g., minimum reporting guidelines, ontology development, or specific FAIR-aligned repositories). The author should propose a minimum information standard for AI-ready nanopharmaceutical datasets, referencing initiatives like MIBBI or NanoMI. 

 

Response: We thank the reviewer for this important and practical suggestion. We agree that the manuscript should provide more actionable recommendations for improving data standardization and AI-readiness in nanopharmaceutical datasets. Accordingly, we revised the manuscript to propose a minimum information framework for AI-ready nanopharmaceutical datasets, aligned with FAIR principles and bio–nano minimum reporting recommendations. The revised text specifies the types of information that should be reported, including nanocarrier composition, synthesis and processing parameters, batch information, physicochemical characterization, biological assay conditions, dose metrics, pharmacokinetic and toxicity endpoints, metadata, model inputs and outputs, validation design, uncertainty estimates, and applicability domain. We also emphasized the need for controlled vocabularies, ontology development, interoperable repositories, and MIBBI-like minimum information standards to improve reproducibility, comparability, and model generalizability. 

 

6. The manuscript stated that many models lack external validation. Similarly, this article does not systematically document how many reviewed studies actually performed external validation or cross-laboratory testing, nor does it propose minimum validation standards. The paper should report a rough estimate of the proportion of studies with external validation and recommend a tiered validation framework. 

 

Response: We thank the reviewer for this important suggestion. We agree that the manuscript should more clearly document the limited extent of external validation in AI-driven nanopharmaceutical studies and provide practical validation recommendations. Because this review is narrative rather than a systematic meta-analysis, we avoided presenting a definitive quantitative estimate for the entire literature. However, we added a qualitative assessment based on the representative studies summarized in Table 2. This analysis indicates that approximately three of the five selected examples incorporated validation beyond retrospective or internal prediction, such as experimental feedback loops, in vivo biological validation, or comparison with independent pharmacokinetic profiles, whereas formulation- or descriptor-based models often still rely on literature-derived datasets, internal validation, curated datasets, or graphical comparison. We also added a tiered validation framework, ranging from internal validation to external validation, prospective experimental validation, and translational validation. This revision clarifies the current validation gap and provides minimum validation standards for future AI-driven nanopharmaceutical studies. 

7. The author mentioned UQ as desirable. But, it does not explain how UQ should be implemented (e.g., Bayesian neural networks, conformal prediction, ensemble variance) or why its absence leads to overconfident and unsafe predictions. The author should include a dedicated subsection to practical UQ methods and their regulatory relevance. 

 

Response: We thank the reviewer for this important suggestion. We agree that uncertainty quantification should be discussed more explicitly, not only as a desirable feature but as a practical requirement for safe and reliable AI implementation in nanopharmaceutical development. Accordingly, we added a dedicated subsection entitled “Validation, Uncertainty Quantification, and Regulatory Relevance.” In this subsection, we explain why the absence of uncertainty estimates may lead to overconfident predictions, particularly when models are applied to formulations, nanocarrier platforms, biological systems, or manufacturing conditions that differ from the training data. We also added practical uncertainty quantification approaches, including Bayesian models, Bayesian neural networks, ensemble methods, Monte Carlo dropout, Gaussian process models, conformal prediction, and applicability-domain analysis. Finally, we clarified that uncertainty quantification should be reported together with model performance metrics, validation strategy, intended use, lifecycle management procedures, and revalidation criteria, especially when AI models are intended to support process control, quality decisions, pharmacokinetic extrapolation, safety assessment, or patient-specific recommendations. 

 

8. The review does not adequately address the absence of prospective clinical validation, the high cost of multimodal patient data, or the ethical and privacy barriers. The manuscript should reframe precision nanomedicine as a long-term goal with major unresolved hurdles, not as a near-term, and recommend specific pilot study designs. 

 

Response: We thank the reviewer for this important and constructive comment. We agree that AI-driven precision nanomedicine should be framed as a long-term translational goal rather than as a near-term clinical application. Accordingly, we revised the manuscript to more clearly emphasize the absence of prospective clinical validation, the high cost and complexity of generating multimodal patient datasets, and the ethical, privacy, governance, and bias-related barriers associated with patient-specific AI models. We also added specific recommendations for staged pilot study designs, including retrospective-to-prospective validation studies, prospective observational biomarker studies, feasibility studies embedded in early-phase nanomedicine trials, and privacy-preserving multicenter studies using federated or distributed learning approaches. These revisions clarify that AI-driven precision nanomedicine remains promising but requires representative patient cohorts, robust consent procedures, secure data governance, bias assessment, prospective validation, and evidence of clinical utility before routine implementation. 

 

Remark Author have tried to fit a broader range of topics to single article, which is commendable but this dilutes the impact of the article. There are not statistical data or experimental outcomes discussed in the article. 

 

Response: We thank the reviewer for this important overall remark. We agree that the manuscript covers a broad range of topics, and we recognize that this breadth may reduce the perceived focus if the relative maturity and practical impact of each AI application are not clearly distinguished. To address this issue, we revised the manuscript to provide a clearer hierarchy of AI applications across the nanopharmaceutical pipeline. Specifically, we now distinguish areas with stronger experimental or preclinical support, such as formulation-property prediction, machine learning-guided nanoparticle optimization, protein corona-based biological fate prediction, and AI-assisted pharmacokinetic modeling, from more exploratory or long-term applications, such as fully autonomous closed-loop development, patient-specific digital twins, reinforcement learning-based process control, and clinical decision-support systems for precision nanomedicine. We also strengthened the manuscript by adding practical examples from the literature and by reporting quantitative performance metrics when available. For example, we now discuss reported R² and RMSE values for AI-assisted pharmacokinetic modeling and comparative findings showing improved predictive performance of machine learning models relative to classical design-of-experiments models in PLGA nanoparticle optimization. In addition, Table 2 was added to summarize representative practical applications, validation strategies, and reported outcomes. We also clarified that the review is narrative rather than a statistical meta-analysis and that performance metrics are not yet reported in a standardized manner across nanopharmaceutical AI studies. These revisions improve the focus of the manuscript and provide readers with a clearer understanding of which AI applications are currently more mature, which remain speculative, and where stronger experimental validation is still needed.

 

Reviewer 4 Report

Comments and Suggestions for Authors

While the scope of this review on advanced drug delivery and pharmaceutical machine learning is highly relevant, the manuscript cannot be accepted for publication in its current form due to significant concerns regarding text originality.

An evaluation of the manuscript indicates an exceptionally high reliance on automated text-generation tools, with analysis suggesting that more than 70% of the text is AI-generated. Review papers are expected to offer deep, authentic, human-authored synthesis and critical analysis of current literature.

The manuscript is being rejected in its current form. If you choose to resubmit this work, the entire text must be completely rewritten in your own words to ensure authentic scientific contribution and adherence to journal authorship guidelines

If you decide to completely rewrite and resubmit the manuscript, you must also address the following technical and presentation issues:

1.All figures require quality upgrades to meet publication standards. Please refine the text within the graphics to ensure clarity, improve overall readability, and resolve any text-rendering overlaps.

2.In model validation and predictive robustness, the manuscript needs to formally define and address the concept of an "applicability domain." Specifically, you should explicitly state that a supervised learning model loses mathematical validity when extrapolated outside its defined training space. Integrating this point will significantly strengthen your discussion on model transparency, algorithmic reliability, and the responsible deployment of predictive frameworks in nanomedicine.

 

Comments for author File: Comments.pdf

Author Response

Response to Reviewer 4

While the scope of this review on advanced drug delivery and pharmaceutical machine learning is highly relevant, the manuscript cannot be accepted for publication in its current form due to significant concerns regarding text originality.

An evaluation of the manuscript indicates an exceptionally high reliance on automated text-generation tools, with analysis suggesting that more than 70% of the text is AI-generated. Review papers are expected to offer deep, authentic, human-authored synthesis and critical analysis of current literature.

The manuscript is being rejected in its current form. If you choose to resubmit this work, the entire text must be completely rewritten in your own words to ensure authentic scientific contribution and adherence to journal authorship guidelines

If you decide to completely rewrite and resubmit the manuscript, you must also address the following technical and presentation issues:

1.All figures require quality upgrades to meet publication standards. Please refine the text within the graphics to ensure clarity, improve overall readability, and resolve any text-rendering overlaps.

2.In model validation and predictive robustness, the manuscript needs to formally define and address the concept of an "applicability domain." Specifically, you should explicitly state that a supervised learning model loses mathematical validity when extrapolated outside its defined training space. Integrating this point will significantly strengthen your discussion on model transparency, algorithmic reliability, and the responsible deployment of predictive frameworks in nanomedicine.

Response: We thank the reviewer and the editorial team for this important assessment. We acknowledge the concern regarding text originality and understand that review articles must provide an authentic, critical, and author-driven synthesis of the literature. In response, we substantially revised the manuscript to improve originality, authorial voice, critical interpretation, and scientific focus. Particular attention was given to the Introduction, where the text was rewritten to provide a more independent and analytical framing of the field, reduce generic or repetitive phrasing, and clarify the central argument of the review. We also revised multiple sections to better distinguish mature AI applications from more speculative or long-term directions, strengthen the discussion of experimental validation and reported performance metrics, and emphasize the translational limitations of current AI models in nanopharmaceutical development.

We also addressed the technical and presentation issues raised by the reviewer. First, all figures were redesigned to improve visual quality, reduce excessive internal text, avoid text overlap, and provide clearer, publication-oriented schematics. The revised figures now use simplified workflows, concise labels, improved readability, and more consistent visual organization. The figure legends were also revised to carry the explanatory content that was previously embedded within the graphics.

Second, we added a more explicit discussion of applicability domain in the section on model validation, uncertainty quantification, and regulatory relevance. We now define applicability domain as the descriptor space, formulation range, experimental conditions, nanocarrier classes, and biological contexts represented in the training data. We explicitly state that a supervised learning model is mathematically reliable only within this domain and that predictions outside the training space represent extrapolation rather than interpolation. We further clarify that out-of-domain predictions may become unreliable, overconfident, or scientifically misleading, and that AI models intended for nanomedicine should identify out-of-domain inputs, report uncertainty, and require additional experimental validation before being used for formulation selection, process control, biological prediction, or regulatory decision-making.

Overall, these revisions were made to strengthen the manuscript’s originality, improve the quality of the figures, and provide a more rigorous discussion of model validity, predictive robustness, and responsible deployment of AI frameworks in nanopharmaceutical development.

 

Reviewer 5 Report

Comments and Suggestions for Authors Gonçalves’s manuscript on the role of artificial intelligence in nanopharmacological development provides a comprehensive overview of the current state of the art in pharmaceutical science. Although the outline is scientifically sound and practically flawless, the text contains some major and minor shortcomings and therefore requires a thorough revision.   - Abstracts divided into sections are not uncommon, though the current format is unusual. Please reorganize the Abstract so that the highlighted keywords appear in separate paragraphs, as indicated by the keywords.   - Why does the author assume that including the terms “artificial intelligence,” “nanopharmacology,” and “clinical application” as keywords — which are already in the title — provides value to the manuscript?   - The Conclusions section requires significant revision. While a few relevant citations may be appropriate in this section, it should contain a summary of the manuscript and the author’s own concluding remarks, rather than ideas or sources cited from elsewhere.   - The References section does not meet the journal’s requirements. The reviewer recommends that the author consult the latest issue of the journal Pharmaceutics or the MDPI website for the relevant guidelines. The reviewer would also like to draw the author’s attention to the fact that references to websites must include the URL, the date of last access, and, if applicable, any information about updates valid at the time of the visit.

Author Response

Response to Reviewer 5

Gonçalves’s manuscript on the role of artificial intelligence in nanopharmacological development provides a comprehensive overview of the current state of the art in pharmaceutical science. Although the outline is scientifically sound and practically flawless, the text contains some major and minor shortcomings and therefore requires a thorough revision.   - Abstracts divided into sections are not uncommon, though the current format is unusual. Please reorganize the Abstract so that the highlighted keywords appear in separate paragraphs, as indicated by the keywords.   - Why does the author assume that including the terms “artificial intelligence,” “nanopharmacology,” and “clinical application” as keywords — which are already in the title — provides value to the manuscript?   - The Conclusions section requires significant revision. While a few relevant citations may be appropriate in this section, it should contain a summary of the manuscript and the author’s own concluding remarks, rather than ideas or sources cited from elsewhere.   - The References section does not meet the journal’s requirements. The reviewer recommends that the author consult the latest issue of the journal Pharmaceutics or the MDPI website for the relevant guidelines. The reviewer would also like to draw the author’s attention to the fact that references to websites must include the URL, the date of last access, and, if applicable, any information about updates valid at the time of the visit. 

Response: We thank the reviewer for the careful evaluation of the manuscript and for recognizing the scientific relevance and overall structure of the review. We have revised the manuscript according to the reviewer’s comments.

First, we reorganized the Abstract to follow a more conventional review-style format. Specifically, we removed the internal section labels and rewrote the Abstract as a continuous summary that presents the background, scope, main findings, limitations, and concluding perspective in a more fluent manner. This revision improves readability and aligns the Abstract more closely with the format commonly used in published review articles.

Second, we revised the Keywords section. We agree that keywords should complement the title rather than repeat its main terms. Therefore, we removed terms already present in the title, such as “artificial intelligence,” “nanopharmaceuticals,” and “clinical translation.” The revised keywords now emphasize more specific technical and translational aspects covered in the manuscript, including machine learning, deep learning, drug delivery, nano–bio interactions, quality-by-design, process analytical technology, digital twins, pharmacokinetics, toxicity prediction, and regulatory science.

Third, we substantially revised the Conclusions section. We agree that this section should primarily summarize the manuscript’s central arguments and provide the author’s own concluding interpretation, rather than relying heavily on cited material. Accordingly, we reduced citation dependency and rewrote the Conclusions to synthesize the main message of the review: AI should be interpreted as a translational framework that connects formulation science, biological prediction, manufacturing control, and clinical decision-making, while its practical impact will depend on standardized datasets, external validation, uncertainty quantification, applicability-domain definition, mechanistic understanding, and clinically meaningful evidence.

Finally, we revised the References section to better follow the journal’s formatting requirements. References were checked for author order, journal abbreviation, year, volume, page range or article number, and DOI where available. Institutional and website-based references were also revised to include the corresponding online source and date of last access, as requested. We did not manually include platform-generated links such as Google Scholar, CrossRef, or PubMed in the manuscript file, as these are typically added by the publisher during online publication.