GCAS: An Integrated R Package and Shiny App for Comprehensive Cancer Data Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Absence of distinct uniqueness and positioning of GCAS: Despite GCAS being portrayed as a comprehensive platform, the book fails to convincingly illustrate its significant superiority over current tools like GEO2R, GEPIA, or other Shiny-based pipelines.A direct benchmarking or comparative investigation is required.

Inadequate validation of GCAS platform efficacy: The paper used NDUFS1 as a case study; however, there is a lack of rigorous validation demonstrating the reproducibility, robustness, or accuracy of GCAS across independent datasets.

Excessive dependence on correlation analysis: Numerous results (e.g., immunological infiltration, drug sensitivity, STAT5/CDKN1A pathway) are predominantly founded on correlation rather than causality.
Functional validation is constrained and fails to comprehensively substantiate mechanistic assertions.

Mechanistic pathway inadequately substantiated: The suggested NDUFS1–STAT5–CDKN1A axis is deficient in comprehensive mechanistic data, including ChIP tests, rescue trials, or pathway blockage studies.

Insufficient statistical rigor: The paper is deficient in details regarding:
Correction for multiple testing (FDR/adjusted p-values)
Effect magnitude and confidence intervals
Statistical power analysis
This undermines the trustworthiness of extensive analyses.

Inadequate attention to heterogeneity within GEO datasets: While a random-effects model is referenced, batch effects, dataset bias, and clinical heterogeneity are inadequately addressed or mitigated.

Restricted biological validation: Experiments are confined to A549 and H1299 cell lines, hence constraining generalizability. No in vivo validation or patient-derived models are incorporated.

Drug sensitivity analysis is devoid of experimental validation. Predictions generated by OncoPredict lack experimental validation for the majority of medications, with the exception of limited data on BI2536, hence diminishing their translational significance.

Ambiguity in the relation of disulfidptosis: Although NDUFS1 is associated with disulfidptosis, the study does not explicitly illustrate disulfidptosis-specific mechanisms or indicators.

Concerns about reproducibility and accessibility: Despite the provision of a Shiny app and a GitHub link, there is an absence of:

Comprehensive user manual
Illustrative datasets
Version control or reproducibility framework

Author Response

Comment1: Absence of distinct uniqueness and positioning of GCAS: Despite GCAS being portrayed as a comprehensive platform, the book fails to convincingly illustrate its significant superiority over current tools like GEO2R, GEPIA, or other Shiny-based pipelines.A direct benchmarking or comparative investigation is required.

Response:

We appreciate this important comment and have substantially revised the manuscript to clarify the positioning of GCAS and to provide a more direct comparison with existing tools.

To address this concern, we added a dedicated subsection in the Methods and a new comparative table (now Table 1) that systematically benchmarks GCAS against widely used platforms, including GEO2R, GEPIA2, UALCAN, and cBioPortal. In this table, we compare the following dimensions: Data sources, Analytical functionalities, Working modes and deployment, Extensibility and customization, Reproducibility support.

We do not claim that GCAS replaces all existing tools; instead, we position it as a complementary, scriptable, and integrative platform particularly suited for researchers who routinely download and integrate GEO/TCGA-like datasets and need both interactive and programmable workflows.

We hope these additions sufficiently clarify the distinct positioning and strengths of GCAS relative to existing tools.

 

 

 

Comment2: Inadequate validation of GCAS platform efficacy: The paper used NDUFS1 as a case study; however, there is a lack of rigorous validation demonstrating the reproducibility, robustness, or accuracy of GCAS across independent datasets.

Response:

We acknowledge that our original emphasis on NDUFS1 as a single, mechanistically oriented case study might have obscured the primary purpose of the manuscript, which is to introduce and validate the GCAS platform. To address this concern and avoid confusion between method development and complex biological mechanistic claims, we made two major changes:

Replacement of the primary case study (NDUFS1 → GAPDH)
In the revised manuscript, we adopt an alternative version of our case study that uses Glyceraldehyde‑3‑phosphate dehydrogenase (GAPDH) instead of NDUFS1 as the exemplar gene.

The rationale is as follows:

• GAPDH is a classical glycolytic enzyme widely used as a housekeeping gene, and its aberrant overexpression in multiple cancers has been consistently reported at both the RNA[1] and protein[2]levels.
• In our previous work, GAPDH has already been examined and validated at both the transcriptional and protein expression levels across multiple cohorts, providing a solid biological and technical foundation.
• Because GAPDH is a well-characterized, robustly measured gene with extensive independent evidence from the literature and from our prior analyses, it is more suitable as a “benchmark-like” example to demonstrate GCAS’s ability to:
  – reproduce known patterns (e.g., overexpression in several tumor types),
  – integrate multiple datasets, and
  – link gene expression to clinical and functional readouts.

We have clarified this in the revised manuscript.

1. Wang, J., et al., GAPDH: A common housekeeping gene with an oncogenic role in pan-cancer. Computational and Structural Biotechnology Journal, 2023. 21: p. 4056-4069. DOI: https://doi.org/10.1016/j.csbj.2023.07.034.
2. Wang, J., et al., PCAS: An Integrated Tool for Multi-Dimensional Cancer Research Utilizing Clinical Proteomic Tumor Analysis Consortium Data. Int J Mol Sci, 2024. 25(12). DOI: 10.3390/ijms25126690.

 

 

Comment3: Excessive dependence on correlation analysis: Numerous results (e.g., immunological infiltration, drug sensitivity, STAT5/CDKN1A pathway) are predominantly founded on correlation rather than causality.

Response:

We fully agree that correlation-based analyses cannot, by themselves, establish causal relationships. In the revised manuscript, we have addressed this concern in two ways.

Methodologically, we:

• explicitly describe the correlation analyses in the Methods section as association studies, and we systematically use cautious wording such as “is associated with” or “correlates with” instead of “regulates” or “drives”;
• emphasize that modules such as immune infiltration, drug-sensitivity prediction, and co‑expression/network analysis in GCAS are designed as hypothesis‑generating tools that highlight potential relationships for further experimental validation, rather than as methods for causal inference.

Textually, we have revised the Results and Discussion to:

• tone down causal language, especially for immunological infiltration and drug‑sensitivity analyses;
• clearly state that these findings should be interpreted as associations that suggest testable hypotheses;
• remove or soften statements implying direct regulation, replacing them with phrasing that reflects the correlational nature of the data.

We believe these changes better align the interpretation of our results with the underlying statistical approach and with the intended role of GCAS as a hypothesis‑generating platform.

 

 

Comment 4: Functional validation is constrained and fails to comprehensively substantiate mechanistic assertions.

Response:

We appreciate this critique and agree that the experimental validation in the original NDUFS1-focused version was not sufficient to fully support strong mechanistic conclusions.

In the revised manuscript, we have altered the emphasis accordingly:

• We have removed the detailed NDUFS1‑centered mechanistic axis from the main narrative and instead adopted a GAPDH‑based case study that focuses on demonstrating GCAS functionality (e.g., differential expression, multi‑cohort integration, survival analysis, pathway enrichment), rather than proposing a complex new pathway.
• We explicitly acknowledge that the functional experiments presented in this work are limited in scope and are intended primarily to illustrate that GCAS‑derived hypotheses can be followed up experimentally, rather than to provide exhaustive mechanistic validation for any single gene.

In the Discussion, we now clearly state that a comprehensive mechanistic characterization would require a separate, dedicated study with extensive in vitro and in vivo experiments, which lies beyond the primary goal of this methods‑oriented manuscript. Our main objective here is to present GCAS and demonstrate its utility in generating biologically plausible and literature‑consistent hypotheses.

 

 

Comment 5: Mechanistic pathway inadequately substantiated: The suggested NDUFS1–STAT5–CDKN1A axis is deficient in comprehensive mechanistic data, including ChIP tests, rescue trials, or pathway blockage studies.

Response:

We appreciate this important comment and agree that the previously proposed NDUFS1–STAT5–CDKN1A axis was not supported by sufficient mechanistic evidence to justify strong causal claims. In the revised manuscript, we have addressed this concern in two major ways.

First, we have changed the focus of the case study from NDUFS1 to GAPDH. GAPDH is a well‑characterized glycolytic enzyme whose overexpression and functional relevance in cancer have been extensively reported and partially validated in our previous work. In this manuscript, the analyses centered on GAPDH are presented as a complement and validation of existing findings (for example, previously reported regulatory relationships such as FOXM1–GAPDH), rather than as a proposal of an entirely novel mechanistic axis. Thus, the current study uses GAPDH primarily to demonstrate how GCAS can systematically integrate and extend prior knowledge, not to claim a fully established new pathway.

Second, based on GCAS‑driven integrative analyses, we identified a putative regulatory relationship between GAPDH and the m6A reader protein IGF2BP3. GCAS suggested that GAPDH expression might be influenced by IGF2BP3, and we followed up this prediction with targeted experimental validation. In the revised manuscript, we now include data supporting that IGF2BP3 positively regulates GAPDH, thereby providing an example of how GCAS‑generated hypotheses can be experimentally confirmed.

 

 

Comment 6: Insufficient statistical rigor: The paper is deficient in details regarding:
Correction for multiple testing (FDR/adjusted p-values)
Effect magnitude and confidence intervals
Statistical power analysis
This undermines the trustworthiness of extensive analyses.

Response:

We thank the reviewer for pointing out the need for more complete statistical reporting. In the revised version, we have addressed these issues at both the methodological description and implementation levels:

1. Multiple testing correction (FDR/adjusted P-values)
   1. All high‑dimensional analyses in GCAS (including differential expression, correlation screening, and enrichment analyses) now explicitly apply Benjamini–Hochberg FDR correction by default.
   2. The corresponding R functions and the Shiny app have been updated so that both raw P‑values and FDR‑adjusted P‑values (e.g., p.adj) are included in all output tablesand used for significance filtering.
   3. These details are now described in the Methods section (Statistical framework of GCAS).
2. Effect sizes and confidence intervals
   1. For differential expression, GCAS outputs log2 fold change (logFC)as the primary effect size, together with P‑values and FDR‑adjusted P‑values.
   2. For correlation analyses, GCAS reports the correlation coefficient (r)and, where applicable, 95% confidence intervals.
   3. For survival analyses, GCAS now reports hazard ratios (HR)with 95% confidence intervals and P‑values.
   4. These effect size and interval estimates have been added to the R function outputs and are visible in the updated Shiny result tables.

We believe these changes improve the statistical transparency and rigor of GCAS, and the revised manuscript now clearly documents these updates in both the R package outputs and the Shiny application.

 

 

Comment 7: Inadequate attention to heterogeneity within GEO datasets: While a random-effects model is referenced, batch effects, dataset bias, and clinical heterogeneity are inadequately addressed or mitigated.

Response:

We appreciate this valuable comment and have clarified and strengthened how GCAS handles heterogeneity across GEO datasets.

In the revised manuscript, we now explicitly describe the multi‑dataset integration module of GCAS, which provides three complementary strategies to address technical and study‑level heterogeneity:

1. Simple intersection analysis across datasets
   1. GCAS allows users to perform independent analyses in each GEO dataset and then identify overlapping differentially expressed genes or signaturesacross studies.
   2. This intersection approach naturally emphasizes signals that are consistent across heterogeneous cohorts, thereby reducing the impact of dataset‑specific bias.
2. Robust Rank Aggregation (RRA)–based integration
   1. GCAS implements RRA analysisto integrate ranked gene lists (e.g., by differential expression) from multiple GEO datasets.
   2. RRA is designed to be robust to outliers and study‑specific noise, providing a consensus ranking that accounts for variability between datasets.
3. ComBat‑based merged analysis with batch correction
   1. When users choose to merge expression matrices from multiple GEO datasets, GCAS applies ComBat(from the package) to correct for batch effects arising from different platforms or studies, while aiming to preserve underlying biological variation.sva
   2. After ComBat adjustment, downstream analyses (e.g., differential expression, clustering) are performed on the batch‑corrected integrated data.

These functionalities are available both in the R package and through the Shiny interface, and are now described in more detail in the Methods section. Together, they provide users with practical tools to mitigate batch effects, reduce dataset‑specific bias, and leverage concordant signals across heterogeneous GEO cohorts. We also note in the Discussion that, despite these measures, residual clinical and biological heterogeneity may remain, and results should be interpreted in that context.

 

 

Comment 8: Restricted biological validation: Experiments are confined to A549 and H1299 cell lines, hence constraining generalizability. No in vivo validation or patient-derived models are incorporated.

Response:

We appreciate this important point regarding biological validation. In the revised manuscript, we have clarified that the primary aim of this work is to introduce and validate GCAS as an integrative cancer transcriptomic analysis platform, and that the experimental data are intended as illustrative examples rather than exhaustive validation across all model systems.

We fully agree that limiting functional assays to A549 and H1299 cell lines constrains generalizability, and that in vivo and patient‑derived models would be required to comprehensively validate any specific mechanistic hypothesis. To address this:

• We have tempered our conclusions, explicitly stating in the Discussion that the current experimental validation is limited to two lung cancer cell lines and that extension to additional cell lines, in vivo models, and patient‑derived samples will be necessary in future work to generalize and refine the biological conclusions.
• We emphasize that GCAS is designed as a hypothesis‑generating platform, and that the present in vitro experiments serve primarily to show that GCAS‑derived hypotheses (e.g., involving GAPDH) can be followed up experimentally, rather than to claim definitive mechanistic proof or clinical translatability.

We plan to explore broader validation, including in vivo and patient‑derived models, in a separate, biology‑focused follow‑up study.

 

 

Comment 9: Drug sensitivity analysis is devoid of experimental validation. Predictions generated by OncoPredict lack experimental validation for the majority of medications, with the exception of limited data on BI2536, hence diminishing their translational significance..

Response:

We agree that the absence of broad experimental validation limits the immediate translational impact of the drug sensitivity analyses. In this manuscript, the OncoPredict‑based module in GCAS is intended as a computational screening tool to prioritize candidate drugs for further testing, rather than as a definitive predictor of clinical response.

To address this concern, we have made the following changes and clarifications:

• In the Methods and Discussion, we now explicitly describe the drug sensitivity analyses as hypothesis‑generating and clearly state that the predicted responses require experimental and/or clinical validation before any therapeutic implications can be drawn.
• We have softened the language around drug sensitivity throughout the Results and Discussion, avoiding any overstatement of translational significance and instead framing the findings as candidate predictions that can help guide subsequent in vitro or in vivo drug testing.

Future work will focus on systematically validating a subset of GCAS‑derived drug predictions in multiple cell lines and, where feasible, in patient‑derived models, but this lies beyond the scope of the present methods‑oriented manuscript.

 

 

Comment 10: Ambiguity in the relation of disulfidptosis: Although NDUFS1 is associated with disulfidptosis, the study does not explicitly illustrate disulfidptosis-specific mechanisms or indicators.

Response:

The revised manuscript no longer relies on NDUFS1/disulfidptosis as a central narrative. Instead, it uses GAPDH as a representative gene to demonstrate GCAS workflows, and it positions disulfidptosis‑related hypotheses as possible future applications of GCAS that will need to be addressed in separate, mechanistically focused studies.

 

Comment 11: Concerns about reproducibility and accessibility: Despite the provision of a Shiny app and a GitHub link, there is an absence of:

Comprehensive user manual
Illustrative datasets
Version control or reproducibility framework

Response:

We thank the reviewer for this comment. In the revised version, we have strengthened reproducibility and accessibility based on the characteristics of the GCAS R package:

• User guidance: We now provide a concise but complete user guide (vignette) within the R package and on GitHub, describing installation, main functions, and example workflows.
• Illustrative datasets: GCAS includes small built‑in example datasets that allow users to run the core modules and reproduce representative analyses directly.
• Versioning and reproducibility: The package is hosted on GitHub with clear version tags/releases, and the exact GCAS version and key dependencies used in this manuscript are specified in the Methods. Example R scripts/Rmd files are provided to reproduce the main analysis steps.

These additions ensure that users have the necessary documentation, example data, and version information to reproduce and extend the analyses performed with GCAS.

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Your tool is interesting and maybe useful, but more data are needed to undercover the difference between yours and other tools, in particular You can underscore the point of strenghtness and weakness, maybe adding a new table (I suggest to add it in discussion).

Materials and methods must be reviewed: in fact, they contains the journal guidelines for pubblishing.

Tables are note clear: I suggest to re-write it with less data but more focused.

The role of NDUFS1 in lung cancer, despite being an interesting field of research, deserve a specific paper and may be confounding for redears: I would delete it and focus only on explain your tool.

Results are clear, but a little scattered and redundant: I suggest to re-write it in a more concise manner.

 

Author Response

Comment 1: Your tool is interesting and maybe useful, but more data are needed to undercover the difference between yours and other tools, in particular You can underscore the point of strenghtness and weakness, maybe adding a new table (I suggest to add it in discussion).

Response:

We thank the reviewer for this helpful suggestion. In the revised manuscript, we have:

• Added a new comparative table (now Table X, referenced in the Discussion) that systematically contrasts GCAS with commonly used tools such as GEO2R, GEPIA2, UALCAN, and cBioPortal.
• The table summarizes and compares:

1. data sources (GEO‑centered vs TCGA/GTEx vs other large cohorts),
2. analytical functionalities (DE, enrichment, immune infiltration, co‑expression/network, drug‑sensitivity prediction, survival analysis, multi‑dataset integration),
3. working modes (web‑only vs scriptable R package + Shiny),
4. extensibility (support for user‑uploaded data and custom pipelines),
5. reproducibility support (versioning, example scripts, example datasets).

In the Discussion, we explicitly highlight both strengths and limitations of GCAS relative to these tools, clarifying that GCAS is positioned as a GEO‑centered, scriptable R package with an integrated Shiny interface, designed to integrate multiple analytic modules and to facilitate reproducible, local workflows.

 

Comment 2: Materials and methods must be reviewed: in fact, they contains the journal guidelines for pubblishing.

Response:

We apologize for this oversight. In the revised manuscript:

We have completely removed any residual journal guideline text from the Materials and Methods section.

The Methods have been thoroughly revised and rewritten, now focusing solely on:

1. data sources and preprocessing,
2. the design and implementation of GCAS (modules, statistical framework, multi‑dataset integration),
3. the details of the case study based on GAPDH,
4. software availability and reproducibility information.

This revision ensures that the Materials and Methods section contains only relevant methodological content and conforms to the journal’s format.

 

Comment 3: Tables are note clear: I suggest to re-write it with less data but more focused.

Response:

We appreciate this comment on clarity. In response: We have simplified and refocused all tables, reducing excessive detail and emphasizing the most relevant information.

 

The role of NDUFS1 in lung cancer, despite being an interesting field of research, deserve a specific paper and may be confounding for redears: I would delete it and focus only on explain your tool.

Response:

We fully agree with the reviewer’s assessment. In the revised manuscript:

We have removed NDUFS1 as the main case study and no longer present the NDUFS1–STAT5–CDKN1A axis as a central theme.

Instead, we now use GAPDH as the representative gene to illustrate GCAS functionality. GAPDH is a well‑studied glycolytic enzyme with established overexpression and functional relevance in multiple cancers, and its role has been partially validated in our previous work.

In this revised version, the GAPDH‑based analyses are framed explicitly as:

a supplement and validation to existing GAPDH literature (e.g., previously reported regulatory relationships such as FOXM1–GAPDH), and a demonstration of GCAS workflows (multi‑dataset integration, differential expression, survival analysis, co‑expression, enrichment, and hypothesis generation).

By doing so, we keep the focus squarely on explaining and validating GCAS as a tool, while using GAPDH as a clear, well‑understood example, and avoid potential confusion arising from an extensive, mechanistically complex NDUFS1 case that would be more appropriate for a separate, biology‑focused paper.

 

Comment 4: Results are clear, but a little scattered and redundant: I suggest to re-write it in a more concise manner.

Response:

We appreciate this observation and have restructured the Results section accordingly:

The Results are now organized to follow the logical workflow of GCAS (data input and preprocessing → single‑dataset analyses → multi‑dataset integration → case‑study demonstration), reducing back‑and‑forth between topics.

We have removed redundant descriptions, merged overlapping subsections, and condensed figures and text to focus on the key outputs that illustrate GCAS capabilities.

We have also toned down mechanistic speculation and correlation‑based interpretations, especially after shifting the case study to GAPDH..

We hope that these revisions address the reviewer’s concerns and make the Results more concise, coherent, and aligned with the manuscript’s primary purpose of presenting the GCAS platform.

Reviewer 3 Report

Comments and Suggestions for Authors

Reviewer Comments

General Comments

The Authors presents an integrated R/Shiny platform for cancer transcriptomic analysis and a case study investigating the role of NDUFS1 in lung cancer. The topic is interesting and the platform may be useful for the researchers in cancer field. However, several important issues should be addressed before the manuscript can be considered for publication.

Major comments:

• The Authors should compare GCAS with existing tools (e.g., GEO2R, GEPIA2, TIMER, cBioPortal) in order to validate it.
• In Methods, IGF2BP3 and GAPDH rather than NDUFS1 are reported, please, check the editing of manuscript.
• Many conclusions are based mainly on results from correlation analyses. Therefore, sentences such as “NDUFS1 regulates” should be mitigated.
• The proposed YTHDF1/NDUFS1/STAT5/CDKN1A pathway needs real experimental validation. Rescue experiments and additional senescence assays would strengthen the conclusions.
• The statistical methods paragraph lacks sufficient detail regarding sample size, multiple testing correction, and biological replicates.

Minor Comments

Overall, the figure legends are long and should be streamlined.

Terminology should be standardized throughout the manuscript.

The authors should report major information regarding GCAS availability, versioning, and reproducibility.

Author Response

Major comments:

Comment 1: The Authors should compare GCAS with existing tools (e.g., GEO2R, GEPIA2, TIMER, cBioPortal) in order to validate it.

Response:

We appreciate this important suggestion. In the revised manuscript, we have:

Added a comparative table (now Table 1) that systematically compares GCAS with GEO2R, GEPIA2, TIMER, UALCAN, and cBioPortal in terms of:

1. data sources (GEO‑centered vs TCGA/GTEx vs other cohorts),
2. core functionalities (differential expression, enrichment, immune infiltration, co‑expression/network, survival analysis, drug response prediction, multi‑dataset integration),
3. working mode (web‑only vs scriptable R package + Shiny interface),
4. extensibility (support for user‑uploaded data and custom pipelines),
5. reproducibility and versioning.

In the Discussion, we now explicitly summarize GCAS’s positioning: a GEO‑centered, scriptable R package with an integrated Shiny app that combines multiple analysis modules and supports local, reproducible workflows, complementary to existing web‑based tools.

These additions clarify the strengths, limitations, and intended role of GCAS relative to commonly used platforms.

 

Comment 2: In Methods, IGF2BP3 and GAPDH rather than NDUFS1 are reported, please, check the editing of manuscript.

Response:

Thank you for pointing this out. We have carefully revised the manuscript to ensure internal consistency:

• The current version no longer uses NDUFS1 as the main case gene. Instead, we have adopted GAPDHas the representative example, supplemented by analyses involving IGF2BP3 as a potential regulator of GAPDH suggested by GCAS and supported by targeted experiments.
• All remaining references to NDUFS1 in the Methods, Results, and figure legends have been removed or corrected as appropriate.
• The Methods section now consistently describes GAPDH (and IGF2BP3 where relevant) as the focus of the illustrative analyses.

 

Comment 3: Many conclusions are based mainly on results from correlation analyses. Therefore, sentences such as “NDUFS1 regulates” should be mitigated.

Response:

We agree that correlation analyses cannot establish causality. In the revised manuscript:

• We have removed NDUFS1‑centered causal statementsand no longer use NDUFS1 as the main example.
• Throughout the text, including for GAPDH and IGF2BP3, we have systematically softened the language, replacing phrases such as “regulates” or “drives” with “is associated with”, “correlates with”, or “is potentially regulated by”, unless supported by specific experimental evidence.
• In the Discussion, we explicitly state that GCAS is a hypothesis‑generating platformand that most associations identified (e.g., immune infiltration, drug sensitivity, co‑expression) should be interpreted as correlations requiring further mechanistic validation.

These changes align the strength of our conclusions with the nature of the underlying analyses.

 

Comment 4: The proposed YTHDF1/NDUFS1/STAT5/CDKN1A pathway needs real experimental validation. Rescue experiments and additional senescence assays would strengthen the conclusions.

Response:
We fully agree that the originally proposed YTHDF1/NDUFS1/STAT5/CDKN1A axis would require substantial additional experimental work (e.g., rescue experiments, pathway inhibition, senescence assays) to be convincingly established.

In light of this, and to keep the manuscript focused on the methodological contribution:

• We have removed the YTHDF1/NDUFS1/STAT5/CDKN1A pathwayfrom the central narrative and no longer present it as a core conclusion of this work.
• The manuscript now uses GAPDHas the main case study to illustrate GCAS workflows. For GAPDH, we emphasize that:
  1. our analyses mainly extend and validateprevious findings (e.g., FOXM1–GAPDH regulation reported in earlier studies), and
  2. GCAS pointed to a potential regulatory relationship between IGF2BP3 and GAPDH, for which we provide targeted experimental support(rather than constructing a multi‑step mechanistic axis).
• We clearly state in the Discussion that comprehensive mechanistic characterization of any specific pathway would require a separate, dedicated studywith extensive in vitro and in vivo experiments, which lies beyond the scope of this methods‑oriented manuscript.

 

Comment 5: The statistical methods paragraph lacks sufficient detail regarding sample size, multiple testing correction, and biological replicates.-

Response:
We thank the reviewer for highlighting this. The statistical framework has been substantially clarified and the implementation updated:

• We added a dedicated subsection “Statistical framework of GCAS” in the Methods, specifying:

1. the use of Benjamini–Hochberg FDRcorrection for multiple testing in high‑dimensional analyses (differential expression, correlation screening, enrichment);
2. reporting of effect sizes(e.g., log2 fold change, correlation coefficients, hazard ratios) and, where applicable, 95% confidence intervals;
3. explicit reporting of sample sizes (n)for each comparison and model in both the text and the GCAS outputs.

• The GCAS R package and Shiny app have been updated so that all relevant outputs include:

1. raw P‑values and FDR‑adjusted P‑values,
2. effect sizes (logFC, r, HR),
3. confidence intervals for correlation and survival analyses.

• We also clarify the number of biological replicates and sample sizes used in the experimental validation parts, and we note in the Discussion that GCAS is primarily an exploratory, hypothesis‑generating platform; formal power considerations should be addressed when designing subsequent validation experiments.

 

 

Minor Comments

Comment 6: Overall, the figure legends are long and should be streamlined.

Response:

We have revised all figure legends to make them more concise and focused:

• Redundant methodological descriptions were removed from legends and moved to the Methods where appropriate.
• Legends now highlight only the key elements needed to understand each figure (dataset, main analysis, main result), while avoiding repetition.

 

Comment 7: Terminology should be standardized throughout the manuscript.

Response:

 We have carefully edited the manuscript to standardize terminology, including:

• consistent use of “GCAS” for the platform,
• consistent naming of datasets, cohorts, and analysis modules,
• harmonized use of terms such as “tumor/normal”, “case/control”, “GEO dataset”, etc.

This should reduce potential confusion and improve readability.

 

 

Comment 8: The authors should report major information regarding GCAS availability, versioning, and reproducibility.

Response:

We agree that this information is essential. In the revised manuscript, we now include a dedicated “Availability and reproducibility” subsection that provides:

• the GCAS R package and Shiny app locations (GitHub URL and Shiny URL),
• the exact version of GCAS used in this study and the key R/package versions,
• a note that GCAS is maintained under version control with tagged releases,
• the availability of a user vignette and example datasets within the R package, as well as example scripts/R Markdown files to reproduce the main workflows.

These additions ensure that readers can access, install, and reproduce the analyses performed with GCAS.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author change and improve as expected.

Reviewer 2 Report

Comments and Suggestions for Authors

None, the paper is acceptable for publication.

Reviewer 3 Report

Comments and Suggestions for Authors

The Authors have addressed all comments/suggestions and really improved the manuscript.