Author response:
The following is the authors’ response to the original reviews.
Reviewer #1 (Public review):
The authors present MerQuaCo, a computational tool that fills a critical gap in the field of spatial transcriptomics: the absence of standardized quality control (QC) tools for image-based datasets. Spatial transcriptomics is an emerging field where datasets are often imperfect, and current practices lack systematic methods to quantify and address these imperfections. MerQuaCo offers an objective and reproducible framework to evaluate issues like data loss, transcript detection variability, and efficiency differences across imaging planes.
Strengths:
(1) The study draws on an impressive dataset comprising 641 mouse brain sections collected on the Vizgen MERSCOPE platform over two years. This scale ensures that the documented imperfections are not isolated or anecdotal but represent systemic challenges in spatial transcriptomics. The variability observed across this large dataset underscores the importance of using sufficiently large sample sizes when benchmarking different image-based spatial technologies. Smaller datasets risk producing misleading results by over-representing unusually successful or unsuccessful experiments. This comprehensive dataset not only highlights systemic challenges in spatial transcriptomics but also provides a robust foundation for evaluating MerQuaCo's metrics. The study sets a valuable precedent for future quality assessment and benchmarking efforts as the field continues to evolve.
(2) MerQuaCo introduces thoughtful metrics and filters that address a wide range of quality control needs. These include pixel classification, transcript density, and detection efficiency across both x-y axes (periodicity) and z-planes (p6/p0 ratio). The tool also effectively quantifies data loss due to dropped images, providing tangible metrics for researchers to evaluate and standardize their data. Additionally, the authors' decision to include examples of imperfections detectable by visual inspection but not flagged by MerQuaCo reflects a transparent and balanced assessment of the tool's current capabilities.
Weaknesses:
(1) The study focuses on cell-type label changes as the main downstream impact of imperfections. Broadening the scope to explore expression response changes of downstream analyses would offer a more complete picture of the biological consequences of these imperfections and enhance the utility of the tool.
Here, we focused on the consequences of imperfections on cell-type labels, one common use for spatial transcriptomics datasets. Spatial datasets are used for so many other purposes that there are almost endless ways in which imperfections could impact downstream analyses. It is difficult to see how we might broaden the scope to include more downstream effects, while providing enough analysis to derive meaningful conclusions, all within the scope of a single paper. Existing studies bring some insight into the impact of imperfections and we expect future studies will extend our understanding of consequences in other biological contexts.
(2) While the manuscript identifies and quantifies imperfections effectively, it does not propose post-imaging data processing solutions to correct these issues, aside from the exclusion of problematic sections or transcript species. While this is understandable given the study is aimed at the highest quality atlas effort, many researchers don't need that level of quality to compare groups. It would be important to include discussion points as to how those cut-offs should be decided for a specific study.
Studies differ greatly in their aims and, as a result, the impact of imperfections in the underlying data will differ also, preventing us from offering meaningful guidance on how cut-offs might best be identified. Rather, our aim with MerQuaCo was to provide researchers with tools to generate information on their spatial datasets, to facilitate downstream decisions on data inclusion and cut-offs.
(3) Although the authors demonstrate the applicability of MerQuaCo on a large MERFISH dataset, and the limited number of sections from other platforms, it would be helpful to describe its limitations in its generalizability.
In figure 9, we addressed the limitations and generalizability of MerQuaCo as best we could with the available datasets. Gaining deep insight into the limitations and generalizability of MerQuaCo would require application to multiple large datasets and, to the best of our knowledge, these datasets are not available.
Reviewer #2 (Public review):
The authors present MerQuaCo, a computational tool for quality control in image-based spatial transcriptomic, especially MERSCOPE. They assessed MerQuaCo on 641 slides that are produced in their institute in terms of the ratio of imperfection, transcript density, and variations of quality by different planes (x-axis).
Strengths:
This looks to be a valuable work that can be a good guideline of quality control in future spatial transcriptomics. A well-controlled spatial transcriptomics dataset is also important for the downstream analysis.
Weaknesses:
The results section needs to be more structured.
We have split the ‘Transcript density’ subsection of the results into 3 new subsections.
Reviewer #3 (Public review):
MerQuaCo is an open-source computational tool developed for quality control in imagebased spatial transcriptomics data, with a primary focus on data generated by the Vizgen MERSCOPE platform. The authors analyzed a substantial dataset of 641 freshfrozen adult mouse brain sections to identify and quantify common imperfections, aiming to replace manual quality assessment with an automated, objective approach, providing standardized data integrity measures for spatial transcriptomics experiments.
Strengths:
The manuscript's strengths lie in its timely utility, rigorous empirical validation, and practical contributions to methodology and biological discovery in spatial transcriptomics.
Weaknesses:
While MerQuaCo demonstrates utility in large datasets and cross-platform potential, its generalizability and validation require expansion, particularly for non-MERSCOPE platforms and real-world biological impact.
We agree that there is value in expanding our analyses to non-Merscope platforms, to tissues other than brain, and to analyses other than cell typing. The limiting factor in all these directions is the availability of large enough datasets to probe the limits of MerQuaCo. We look forward to a future in which more datasets are available and it’s possible to extend our analyses
Reviewer #1(Recommendation for the Author):
(1) To better capture the downstream impacts of imperfections, consider extending the analysis to additional metrics, such as specificity variation across cell types, gene coexpression, or spatial gene patterning. This would deepen insights into how these imperfections shape biological interpretations and further demonstrate the versatility of MerQuaCo.
These are compelling ideas, but we are unable to study so many possible downstream impacts in sufficient depth in a single study. Insights into these topics will likely come from future studies.
(2) In Figure 7 legend, panel label (D) is repeated thus panels E-F are mislabelled.
We have corrected this error.
(3) Ensure that the image quality is high for the figures.
We will upload Illustrator files, ensuring that images are at full resolution.
Reviewer #2 (Recommendation for the Author):
(1) A result subsection "Transcript density" looks too long. Please provide a subsection heading for each figure.
We have split this section into 3 with new subheadings.
(2) The result subsection title "Transcript density" sounds ambiguous. Please provide a detailed title describing what information this subsection contains.
We have renamed this section ‘Differences in transcript density between MERSCOPE experiments’.
Minor:
(1) There is no explanation of the black and grey bars in Figure 2A.
We have added information to the figure legend, identifying the datasets underlying the grey and black bars.
(2) In the abstract, the phrase "High-dimension" should be "High-dimensional".
We have changed ‘high-dimension’ to ‘high-dimensional’.
(3) In the abstract, "Spatial results" is an unclear expression. What does it stand for?
We have replaced the term ‘spatial results’ with ‘the outputs of spatial transcriptomics platforms’.
Reviewer #3 (Recommendation for the Author):
(1) While the tool claims broad applicability, validation is heavily centered on MERSCOPE data, with limited testing on other platforms. The authors should expand validation to include more diverse platforms and add a small analysis of non-brain tissue. If broader validation isn't feasible, modify the title and abstract to reflect the focus on the mouse brain explicitly.
We agree that expansion to other platforms is desirable, but to the best of our knowledge sufficient datasets from other platforms are not available. In the abstract, we state that ‘… we describe imperfections in a dataset of 641 fresh-frozen adult mouse brain sections collected using the Vizgen MERSCOPE.’
(2) The impact of data imperfections on downstream analysis needs a more comprehensive evaluation. The authors should expand beyond cluster label changes to include a) differential expression analysis with simulated imperfections, b) impact on spatial statistics and pattern detection, and c) effects on cell-cell interactions.
Each of these ideas could support a substantial study. We are unable to do them justice in the limited space available as an addition to the current study.
(3) The pixel classification workflow and validation process need more detailed documentation.
The methods and results together describe the workflow and validation in depth. We are unclear what details are missing.
(4) The manuscript lacks comparison to existing. QC pipelines such as Squidpy and Giotto. The authors should benchmark MerQuaCo against them and provide integration options with popular spatial analysis tools with clear documentation.
To the best of our knowledge, Squidpy and Giotto lack QC benchmarks, certainly of the parameters characterized by MerQuaCo. Direct comparison isn’t possible.
