Calibrated analysis framework for nanopore direct RNA sequencing uncovers cell-specific m⁶A proportions at conserved sites
Curation statements for this article:-
Curated by eLife
eLife Assessment
This valuable study compares orthogonal approaches for detecting RNA chemical modifications and provides a helpful framework for improving the reliability of RNA modification identification by direct RNA sequencing. The evidence supporting the technical benchmarking claims is solid, with multiple complementary controls, including in vitro-transcribed RNA, chemical inhibition of METTL3, and orthogonal m6A mapping. However, claims regarding relative method accuracy and cell-type- and isoform-specific biological differences remain less well supported due to limited biological replication and validation.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Nanopore direct RNA sequencing (DRS) coupled with Dorado modification-aware basecalling enables mapping of epitranscriptomic modifications including N6-methyladenosine (m6A) at the level of individual RNAs. However, the sensitivity, specificity, and reproducibility of this method remain unclear and have only recently begun to be addressed through systematic benchmarking studies. Here, we aimed to establish a best-practice workflow for DRS-based epitranscriptomic analyses. Specifically, we evaluated multiple Dorado versions and models using RNA isolated from primary cells and unmodified in vitro transcribed RNAs. We further utilized an m6A methyltransferase inhibitor as a specificity control. We established that stringent filtering is necessary to reduce the contribution of background and found that Dorado predictions captured an increasing proportion of GLORI sites detected at high m6A/A proportions. Further, by applying DRS to a human primary fibroblast and a HD10.6 neuron dataset, we detected differences in the predicted m6A/A proportions at conserved sites between these datasets. Our study thus presents the first systematic comparison of Dorado and GLORI from the same input RNA and expands characterization of the m6A epitranscriptome to fibroblasts and neurons.
Article activity feed
-
eLife Assessment
This valuable study compares orthogonal approaches for detecting RNA chemical modifications and provides a helpful framework for improving the reliability of RNA modification identification by direct RNA sequencing. The evidence supporting the technical benchmarking claims is solid, with multiple complementary controls, including in vitro-transcribed RNA, chemical inhibition of METTL3, and orthogonal m6A mapping. However, claims regarding relative method accuracy and cell-type- and isoform-specific biological differences remain less well supported due to limited biological replication and validation.
-
Reviewer #1 (Public review):
[Editors' note: this version has been assessed by the Reviewing Editor without further input from the original reviewers. The authors have addressed the comments raised in the previous round of review.]
The authors have substantially strengthened the manuscript in response to my comments. In particular, they have improved the treatment of non-canonical calls, added further illustrative examples at the level of individual RNA variants, repeated pathway enrichment using an appropriate expression-matched background, assessed treatment-associated cell viability, and moderated several biological claims. The revised manuscript is now most compelling as a technical benchmarking and best-practice resource for high-confidence prediction of a common RNA chemical modification from nanopore direct RNA sequencing.
The …
Reviewer #1 (Public review):
[Editors' note: this version has been assessed by the Reviewing Editor without further input from the original reviewers. The authors have addressed the comments raised in the previous round of review.]
The authors have substantially strengthened the manuscript in response to my comments. In particular, they have improved the treatment of non-canonical calls, added further illustrative examples at the level of individual RNA variants, repeated pathway enrichment using an appropriate expression-matched background, assessed treatment-associated cell viability, and moderated several biological claims. The revised manuscript is now most compelling as a technical benchmarking and best-practice resource for high-confidence prediction of a common RNA chemical modification from nanopore direct RNA sequencing.
The matched-input comparison of two methods, use of unmodified RNA controls, and inhibitor treatment are important strengths. Together, these experiments provide solid support for the central conclusion that careful filtering and appropriate controls can substantially reduce likely false-positive calls and yield a reproducible set of high-confidence predictions. The added analyses also provide useful illustrations of the capacity of long-read sequencing to assign predictions to individual RNA variants, although they do not establish broad superiority over the short-read comparison method.
-
Reviewer #2 (Public review):
The revised manuscript has improved substantially. The authors have addressed the main concerns raised during the previous round by introducing an IVT-based calibration framework, evaluating background calls across probability thresholds, examining basecalling mismatches, and replacing "stoichiometry" with the more appropriate term "predicted modification proportion." The expanded discussion of non-DRACH predictions is also useful and appropriately balanced. Overall, the study now provides a valuable practical framework for Dorado-based RNA modification analysis.
-
Author response:
The following is the authors’ response to the previous reviews.
Public Reviews:
Reviewer #1 (Public review):
The authors have substantially strengthened the manuscript in response to my comments. In particular, they have improved the treatment of non-canonical calls, added further illustrative examples at the level of individual RNA variants, repeated pathway enrichment using an appropriate expression-matched background, assessed treatment-associated cell viability, and moderated several biological claims. The revised manuscript is now most compelling as a technical benchmarking and best-practice resource for high-confidence prediction of a common RNA chemical modification from nanopore direct RNA sequencing.
The matched-input comparison of two methods, use of unmodified RNA controls, and inhibitor treatment are …
Author response:
The following is the authors’ response to the previous reviews.
Public Reviews:
Reviewer #1 (Public review):
The authors have substantially strengthened the manuscript in response to my comments. In particular, they have improved the treatment of non-canonical calls, added further illustrative examples at the level of individual RNA variants, repeated pathway enrichment using an appropriate expression-matched background, assessed treatment-associated cell viability, and moderated several biological claims. The revised manuscript is now most compelling as a technical benchmarking and best-practice resource for high-confidence prediction of a common RNA chemical modification from nanopore direct RNA sequencing.
The matched-input comparison of two methods, use of unmodified RNA controls, and inhibitor treatment are important strengths. Together, these experiments provide solid support for the central conclusion that careful filtering and appropriate controls can substantially reduce likely false-positive calls and yield a reproducible set of high-confidence predictions. The added analyses also provide useful illustrations of the capacity of long-read sequencing to assign predictions to individual RNA variants, although they do not establish broad superiority over the short-read comparison method.
A few claims remain insufficiently supported. First, the discussion of non-DRACH sites should cite Cancer mutations rewire the RNA methylation specificity of METTL3-METTL14 (Science Advances, DOI: 10.1126/sciadv.ads4750). In particular, Figure 3C of that study shows that the normal METTL3-METTL14 enzyme retains measurable activity on non-DRACH substrates; this provides a useful precedent for describing the inhibitor-sensitive GGAUU calls as plausible candidates rather than treating non-DRACH calls, especially those detected only by GLORI, as presumptively artefactual. The revised manuscript appropriately notes technical and biological explanations for non-canonical calls and identifies the GGAUU sites as candidates requiring further validation, but this additional context would improve interpretation.
We thank the reviewer for reminding us of this interesting publication. We have included the citation and adjusted our discussion accordingly (lines 576-579).
Second, the comparison between Dorado and GLORI remains overinterpreted. The matched-input design and inhibitor perturbation are important strengths, but the current analysis does not establish that sites detected only by GLORI are false positives or that Dorado is a reference standard. This would require, for example, direct comparison of inhibitor responsiveness among GLORI-only and shared sites, while accounting for coverage, conversion efficiency, and other method-specific biases; therefore, phrases such as "GLORI exhibits reduced specificity," "largely spurious," and "validated by an orthogonal approach" are not fully supported by the data presented.
Thank you for this comment. We agree that our original language potentially overinterpreted the GLORI-Dorado comparison. In our GLORI dataset, the relatively high gene-level non-conversion rate (>5%) particularly limits interpretation of low-proportion sites. Although this background has little effect at high m6A/A proportions, it can substantially affect calls in the 10-20% range. Consistent with this limitation, GLORI-only sites were predominantly low-stoichiometry and more often occurred outside DRACH motifs, whereas shared sites were enriched at higher predicted m6 A/A proportions. GLORI coverage was similar between categories, and all sites analyzed were STM2457-responsive. However, residual non-conversion may disproportionately affect low-fraction calls. Thus, our findings do not establish that GLORI-only sites are false positives but instead indicate unresolved method-specific differences. We have adjusted the section accordingly (lines 327-382) and added two additional subplots in Fig. S21 (lines 1193-1202 contain the new figure legends).
Similarly, the thresholding approach should not be described as "unbiased." The analysis is useful and empirically informed, but it relies on selected parameters, a manually specified penalty, and predetermined decision criteria. Describing the procedure as "data-driven" or "empirically calibrated against matched unmodified RNA controls" would more accurately convey both its value and its limitations.
Thank you for this comment. We agree that this language should be used carefully. To our knowledge, we have already refined the relevant wording in the current revised version.
The additional RNA-variant examples are helpful, but they demonstrate an advantage of long-read data rather than broadly proving that Dorado "outperforms" GLORI for RNA-variant-level quantification. A more accurate formulation would be: "Direct RNA sequencing provides RNA-variant-resolved predictions that are not readily assignable using short-read GLORI data, as illustrated at several loci."
We have revised the wording to emphasize that DRS enables RNA-variant-resolved predictions that are not readily assignable using short-read GLORI data (lines 384-385 and 404-406).
Finally, the authors now acknowledge that the functional significance of cell-type differences remains unresolved, but the abstract still presents the results as evidence of "finely tuned" regulation, and the discussion links them to neuronal function. The data provide solid evidence for differences in predicted modification proportions between cell types, but the evidence is incomplete for claims about functional consequences; these statements should therefore be framed as context or hypotheses for future study rather than as inferences from the present experiments.
Thank you for this comment. In the revised manuscript, we do not describe the observed cell-type differences as evidence of “finely tuned” regulation or infer functional consequences for neuronal function. Any potential relevance is framed only as context for future investigation. We refer to lines 43-46 and 602-606, where we describe our observations about increased m6A/A site ratios in HD10.6 cells compared to NHDFs.
Reviewer #2 (Public review):
The revised manuscript has improved substantially. The authors have addressed the main concerns raised during the previous round by introducing an IVT-based calibration framework, evaluating background calls across probability thresholds, examining base calling mismatches, and replacing "stoichiometry" with the more appropriate term "predicted modification proportion." The expanded discussion of non-DRACH predictions is also useful and appropriately balanced. Overall, the study now provides a valuable practical framework for Dorado-based RNA modification analysis.
I have only a few remaining comments, all of which should be addressable through clarification and minor textual revisions.
(1) The comparison between NHDF and HD10.6 cells does not appear to include independent biological replicates. The observed differences in predicted m6A/A proportions should therefore be presented as differences between the two datasets rather than as general cell-type-specific effects. The interpretation is already cautious in the Discussion, but similar wording should be used in the Abstract, Results heading, and title if applicable.
We thank the reviewer for this suggestion and agree that the wording should avoid overinterpretation. We consider the title appropriate, as the manuscript reports differences in predicted m6 A proportions between fibroblast and neuronal datasets. However, to ensure greater precision, we revised the relevant sentence in the Abstract (lines 43-46) and two Results section headings (lines 130-131 and 408-409).
(2) The direction of the comparison in Figure 3D should be clarified. The percentages represent the fraction of GLORI sites that is also detected by Dorado. Thus, the analysis shows that Dorado increasingly recovers high-proportion GLORI sites. It does not directly demonstrate that high-proportion Dorado predictions are increasingly validated by GLORI. The text should reflect this distinction, and the figure should clearly identify which circle represents each method.
We thank the reviewer for this clarification. We revised the text in lines 367-369 to clarify that Dorado increasingly recovers high-proportion GLORI sites, and we adjusted Fig. 3D to improve clarity.
(3) The ratio of calls detected in IVT relative to biological samples is a useful empirical measure of background, but it is not formally a false-positive rate because the true status and testability of every site are not known. A term such as "IVT-to-sample call ratio" or "empirical background-call ratio" would be more precise and would not alter the main conclusion regarding the value of stringent filtering.
The reviewer’s point is well taken, and we have replaced the use of false-positive with ‘background signal’ throughout the manuscript when dealing with our approach and results. We have however retained the use of false-positive as a term when directly quoting from other published studies (regardless of whether they used this term correctly).
(4) The description of the change-point framework still contains the placeholder "at least X-fold more signal." The actual enrichment threshold should be provided. The authors should also ensure that a log2 ratio threshold of 1.5 is not described as 1.5-fold enrichment, since these values are not equivalent.
We thank the reviewer for catching this. We have corrected the sentence on lines 739-740 to note that the fold enrichment is 1.5-fold which equates to a log2 ratio threshold of ~0.585.
These are relatively minor issues. The authors have responded constructively to the previous reviews, and I do not consider further experimental work necessary.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) Add the Science Advances reference, DOI: 10.1126/sciadv. ads4750, to the discussion of non-DRACH sites and use it to contextualize inhibitor-sensitive GGAUU sites as plausible candidates requiring independent validation.
(2) Revise language implying that GLORI-only sites are false positives, that GLORI has reduced specificity, or that Dorado serves as the validating reference method; instead, describe the observed agreement and acknowledge unresolved method-specific differences.
(3) If feasible, compare inhibitor responsiveness directly among shared and GLORI-only sites, reporting coverage and conversion-related quality measures for those categories.
(4) Replace "unbiased" thresholding language with "data-driven" or "empirically calibrated against matched unmodified RNA controls."
(5) Revise broad "outperforms" language about RNA-variant-level quantification to specify that long-read sequencing enables predictions at individual RNA variants in the illustrated examples.
(6) Remove or further qualify statements that infer finely tuned regulation or neuronal functional effects from differences in predicted modification proportions alone.
We have addressed these recommendations as described above.
-
-
-
Calibrated analysis framework for nanopore direct RNA sequencing uncovers cell-specific m6A proportions at conserved sites bioRxiv · Ohnezeit et al. doi.org/10.1101/2025.11.02.686099 #RNAsky #RNAmods
-
eLife Assessment
This valuable study compares orthogonal approaches for detecting RNA chemical modifications and provides a helpful framework for improving the reliability of RNA modification identification by direct RNA sequencing. The evidence supporting the technical benchmarking claims is solid, with multiple complementary controls, including in vitro-transcribed RNA, chemical inhibition of METTL3, and orthogonal m6A mapping. However, claims regarding relative method accuracy and cell-type- and isoform-specific biological differences remain less well supported due to limited biological replication and validation.
-
Reviewer #1 (Public review):
The authors have substantially strengthened the manuscript in response to my comments. In particular, they have improved the treatment of non-canonical calls, added further illustrative examples at the level of individual RNA variants, repeated pathway enrichment using an appropriate expression-matched background, assessed treatment-associated cell viability, and moderated several biological claims. The revised manuscript is now most compelling as a technical benchmarking and best-practice resource for high-confidence prediction of a common RNA chemical modification from nanopore direct RNA sequencing.
The matched-input comparison of two methods, use of unmodified RNA controls, and inhibitor treatment are important strengths. Together, these experiments provide solid support for the central conclusion that …
Reviewer #1 (Public review):
The authors have substantially strengthened the manuscript in response to my comments. In particular, they have improved the treatment of non-canonical calls, added further illustrative examples at the level of individual RNA variants, repeated pathway enrichment using an appropriate expression-matched background, assessed treatment-associated cell viability, and moderated several biological claims. The revised manuscript is now most compelling as a technical benchmarking and best-practice resource for high-confidence prediction of a common RNA chemical modification from nanopore direct RNA sequencing.
The matched-input comparison of two methods, use of unmodified RNA controls, and inhibitor treatment are important strengths. Together, these experiments provide solid support for the central conclusion that careful filtering and appropriate controls can substantially reduce likely false-positive calls and yield a reproducible set of high-confidence predictions. The added analyses also provide useful illustrations of the capacity of long-read sequencing to assign predictions to individual RNA variants, although they do not establish broad superiority over the short-read comparison method.
A few claims remain insufficiently supported. First, the discussion of non-DRACH sites should cite Cancer mutations rewire the RNA methylation specificity of METTL3-METTL14 (Science Advances, DOI: 10.1126/sciadv.ads4750). In particular, Figure 3C of that study shows that the normal METTL3-METTL14 enzyme retains measurable activity on non-DRACH substrates; this provides a useful precedent for describing the inhibitor-sensitive GGAUU calls as plausible candidates rather than treating non-DRACH calls, especially those detected only by GLORI, as presumptively artefactual. The revised manuscript appropriately notes technical and biological explanations for non-canonical calls and identifies the GGAUU sites as candidates requiring further validation, but this additional context would improve interpretation.
Second, the comparison between Dorado and GLORI remains overinterpreted. The matched-input design and inhibitor perturbation are important strengths, but the current analysis does not establish that sites detected only by GLORI are false positives or that Dorado is a reference standard. This would require, for example, direct comparison of inhibitor responsiveness among GLORI-only and shared sites, while accounting for coverage, conversion efficiency, and other method-specific biases; therefore, phrases such as "GLORI exhibits reduced specificity," "largely spurious," and "validated by an orthogonal approach" are not fully supported by the data presented.
Similarly, the thresholding approach should not be described as "unbiased." The analysis is useful and empirically informed, but it relies on selected parameters, a manually specified penalty, and predetermined decision criteria. Describing the procedure as "data-driven" or "empirically calibrated against matched unmodified RNA controls" would more accurately convey both its value and its limitations.
The additional RNA-variant examples are helpful, but they demonstrate an advantage of long-read data rather than broadly proving that Dorado "outperforms" GLORI for RNA-variant-level quantification. A more accurate formulation would be: "Direct RNA sequencing provides RNA-variant-resolved predictions that are not readily assignable using short-read GLORI data, as illustrated at several loci."
Finally, the authors now acknowledge that the functional significance of cell-type differences remains unresolved, but the abstract still presents the results as evidence of "finely tuned" regulation, and the discussion links them to neuronal function. The data provide solid evidence for differences in predicted modification proportions between cell types, but the evidence is incomplete for claims about functional consequences; these statements should therefore be framed as context or hypotheses for future study rather than as inferences from the present experiments.
-
Reviewer #2 (Public review):
The revised manuscript has improved substantially. The authors have addressed the main concerns raised during the previous round by introducing an IVT-based calibration framework, evaluating background calls across probability thresholds, examining basecalling mismatches, and replacing "stoichiometry" with the more appropriate term "predicted modification proportion." The expanded discussion of non-DRACH predictions is also useful and appropriately balanced. Overall, the study now provides a valuable practical framework for Dorado-based RNA modification analysis.
I have only a few remaining comments, all of which should be addressable through clarification and minor textual revisions.
(1) The comparison between NHDF and HD10.6 cells does not appear to include independent biological replicates. The observed …
Reviewer #2 (Public review):
The revised manuscript has improved substantially. The authors have addressed the main concerns raised during the previous round by introducing an IVT-based calibration framework, evaluating background calls across probability thresholds, examining basecalling mismatches, and replacing "stoichiometry" with the more appropriate term "predicted modification proportion." The expanded discussion of non-DRACH predictions is also useful and appropriately balanced. Overall, the study now provides a valuable practical framework for Dorado-based RNA modification analysis.
I have only a few remaining comments, all of which should be addressable through clarification and minor textual revisions.
(1) The comparison between NHDF and HD10.6 cells does not appear to include independent biological replicates. The observed differences in predicted m6A/A proportions should therefore be presented as differences between the two datasets rather than as general cell-type-specific effects. The interpretation is already cautious in the Discussion, but similar wording should be used in the Abstract, Results heading, and title if applicable.
(2) The direction of the comparison in Figure 3D should be clarified. The percentages represent the fraction of GLORI sites that is also detected by Dorado. Thus, the analysis shows that Dorado increasingly recovers high-proportion GLORI sites. It does not directly demonstrate that high-proportion Dorado predictions are increasingly validated by GLORI. The text should reflect this distinction, and the figure should clearly identify which circle represents each method.
(3) The ratio of calls detected in IVT relative to biological samples is a useful empirical measure of background, but it is not formally a false-positive rate because the true status and testability of every site are not known. A term such as "IVT-to-sample call ratio" or "empirical background-call ratio" would be more precise and would not alter the main conclusion regarding the value of stringent filtering.
(4) The description of the change-point framework still contains the placeholder "at least X-fold more signal." The actual enrichment threshold should be provided. The authors should also ensure that a log2 ratio threshold of 1.5 is not described as 1.5-fold enrichment, since these values are not equivalent.
These are relatively minor issues. The authors have responded constructively to the previous reviews, and I do not consider further experimental work necessary.
-
Author response:
The following is the authors’ response to the original reviews.
We thank you all for the careful evaluation of our manuscript and for the numerous constructive comments and suggestions. As noted by the reviewers, the study directly compares two distinct approaches for detecting m6A modifications using the same RNA input material. The inclusion of IVT and STM2457-treated RNA as negative controls, as well as the high-depth long-read datasets from two different cellular systems, enable robust analysis and interpretation of the data, providing a rigorous and reproducible analysis workflow and a valuable reference for the rapidly expanding community aiming to use Oxford Nanopore Technologies’ (ONT) latest flow cell chemistry and Dorado base calling tool.
We have revised the manuscript according to this feedback and believe …
Author response:
The following is the authors’ response to the original reviews.
We thank you all for the careful evaluation of our manuscript and for the numerous constructive comments and suggestions. As noted by the reviewers, the study directly compares two distinct approaches for detecting m6A modifications using the same RNA input material. The inclusion of IVT and STM2457-treated RNA as negative controls, as well as the high-depth long-read datasets from two different cellular systems, enable robust analysis and interpretation of the data, providing a rigorous and reproducible analysis workflow and a valuable reference for the rapidly expanding community aiming to use Oxford Nanopore Technologies’ (ONT) latest flow cell chemistry and Dorado base calling tool.
We have revised the manuscript according to this feedback and believe the changes have greatly improved the clarity and utility of the work. We have, where possible, addressed all the points raised and have further discussed or acknowledged those that involve broader technical considerations beyond the scope of the present study.
Reviewer #1 (Public review):
Summary:
The authors set out to evaluate how accurately direct sequencing of RNA can identify and quantify several chemical modifications on RNA molecules, focusing primarily on m6A. A central goal of the work is to compare this approach with an independent chemical-based method (glyoxal and nitrite-mediated deamination of unmethylated adenosines (GLORI), using the same RNA samples, in order to assess reproducibility, false-positive signals, and sensitivity across a range of detection strategies. The authors further aim to demonstrate the biological utility of this approach by applying it to two human cell types, primary human fibroblasts and HD10.6 neurons. While the manuscript also reports detection of additional RNA modifications (pseudouridine and m5C), the depth of analysis and strength of controls are greatest for m6A, which forms the primary focus of the study.
Strengths:
A strength of this work is the direct comparison of two distinct measurement approaches performed on the same RNA input material; this has not been done in other recently published benchmarking studies evaluating the utility of the recent direct RNA sequencing for calling m6A. The authors systematically test multiple analysis models and show that, when appropriate filtering is applied, detection of modified sites is reproducible across software versions. The use of synthetic RNA standards and METTL3 inhibitors as negative controls helps to reinforce the overall results. The data show good agreement between the two methods at higher m6A modification levels, supporting the conclusion that direct RNA sequencing can reliably detect high-confidence modification sites. The authors also demonstrate that this approach can, in principle, provide information at the level of individual RNA variants (although only one example was provided), which is difficult to achieve with short-read methods. The methodology described here is likely to be useful to others seeking to apply similar approaches to identify and quantify m6A. The study also explores the detection of other RNA modifications, which highlights the broader potential of the approach, although these analyses are necessarily more exploratory given the more limited controls and data available.
Weaknesses:
Despite these strengths, several issues limit the interpretation of the results and should be clarified for readers.
First, the authors appropriately address false-positive signals by estimating expected false-positive rates and by quantitatively comparing sequence motif enrichment before and after filtering. These analyses provide important support for the use of stoichiometry-based thresholds and demonstrate that filtering substantially improves specificity. However, even after filtering, a subset of detected sites remains outside the expected sequence context. It therefore remains unclear to what extent these non-canonical sites reflect genuine biology versus residual technical artifacts.
The detected sites that were outside of the expected sequence context (0.1% of the sites per cell type) exhibited sensitivity to STM2457 treatment, supporting the conclusion that these sites represent genuine signals. This point has been clarified in the text (lines 424-455 and 563-578) and additional supporting data have been included in the supplementary figures (Supplementary Fig. S23 and S24).
Second, claims regarding the ability of direct RNA sequencing to resolve modification patterns across different RNA variants are supported by very limited evidence. The conclusion that this approach provides superior isoform-level quantification relative to short-read methods is based largely on a single gene example. While this case is interesting, it does not establish how widespread or general this advantage is. A broader analysis indicating how many genes show isoform-specific modification patterns detectable by this method, and how often these are missed by the comparison approach, would be necessary to support a general claim.
We thank the reviewer for this valuable comment. To strengthen this point, we have added two additional examples illustrating isoform-specific m6A modification patterns, as well as one monoexonic gene for comparison (new Supplementary Figure S22). KRTAP1-5 is a monoexonic gene with a single detectable isoform and shows that Dorado detects more m6A sites than GLORI. HSPA8, with five detectable isoforms, is a good example to show that two isoforms carry additional m6A sites not resolvable by GLORI. CITED2 has three detectable isoforms, and differential m6A patterns can only be resolved with Dorado. We adjusted the text accordingly in lines 395-401.
Third, the biological interpretation of cell type-specific differences in modification levels remains underdeveloped. Although differences in modification stoichiometry are reported between fibroblasts and neuron-derived cells, the functional consequences of these differences are not addressed. It is unclear whether changes in modification levels are associated with differences in RNA abundance, stability, or translation. As a result, statements suggesting that these modifications fine-tune core cellular pathways are speculative and should either be supported with additional analyses or framed more cautiously.
We have updated the text (lines 477-481) to ensure a more cautious interpretation of the observations.
Related to this point, differences in gene expression between the two cell types are a potential confounding factor. The pathway enrichment patterns presented appear biased toward particular functional categories, but without clear control for differential gene expression, it is difficult to determine whether the observed enrichment reflects cell type-specific regulation of RNA modification or simply differences in which genes are expressed. Clarifying how background gene sets were defined for these analyses would help readers interpret the results.
We thank the reviewer for raising this important point. For the KEGG pathway enrichment analysis, we used genes harboring shared m6A sites in both NHDF and HD10.6 cells as the input gene list, irrespective of their expression levels. We initially did not use a custom background gene set, which is also a valid point brought up by the reviewer. Hence, we repeated the analysis and used all genes expressed in both cell types (~13,000 genes) as background (i.e. excluding those only expressed in one cell type). We have revised the results (lines 467-468) and methods section (lines 740-749) accordingly. We replaced the plot in Figure 4H with the new plot.
The manuscript also suggests broader differences in overall modification levels between cell types, but this is not validated using an independent global assay. An orthogonal measurement of total modification levels on polyadenylated RNA (for example, dot blot) would help place site-specific stoichiometry differences in a clearer biological context.
We tried to assess global changes in modification levels using a dot blot assay. However, even when loading 2 µg of RNA per sample, we did not detect visible differences between DMSO- and STM2457-treated samples in either cell type, which we expected to serve as a positive control. This most likely indicates that the dot blot assay is much less sensitive than our DRS-based sequencing approach. However, as we agree that for making a statement about broader differences between cell types, orthogonal validation would be necessary. Accordingly, we have adjusted our statements in lines 44-46 and 477-481.
Finally, the effects of the METTL3 inhibitor on these cell types are not fully characterized. While changes in m6A modification patterns are reported following treatment, the manuscript does not address whether the treatment affects cell growth or viability.
Cell viability following STM2457 treatment was determined in NHDF and HD10.6 cells using the CellTiter-Glo® assay. HD10.6 cells are terminally differentiated and post-mitotic; therefore, proliferation/growth is not applicable to this cell type, and cell viability (assessed by CTG assay) is the relevant endpoint. We added the results, which showed no significant differences in either cell type, as a supplementary figure (Fig. S1). While we ensured that NHDFs showed normal growth behaviour when treating them with STM2457, because HD10.6 cells are non-proliferative, we assessed cellular health in each of the experiments conducted. We updated a description of the assay in the methods section, lines 637-643, and in the results section (lines 137-138).
Appraisal of conclusions and impact:
Overall, the study provides an informative technical assessment of direct RNA sequencing for modification detection and establishes clear conditions under which the method performs well. The evidence strongly supports conclusions related to technical benchmarking, reproducibility, and the importance of filtering and controls, particularly for m6A. In contrast, conclusions regarding isoform-specific regulation and cell type-specific biological roles of RNA modification are less well supported by the data currently presented, and would benefit from either additional analysis or more restrained interpretation.
The work is likely to have a meaningful impact as a practical reference for researchers using direct RNA sequencing, particularly by clarifying sources of false positives and the value of appropriate controls. With clearer limits placed on biological interpretation or more data presented in support of the biological interpretation, the study would serve as a valuable reference for users seeking to apply these technologies reliably.
We thank the reviewer for this positive evaluation of our work. To address major concerns, we have adjusted our interpretation of the biological findings throughout the manuscript to better reflect the scope of the presented data. We believe these revisions more clearly distinguish the technical benchmarking aspects of the study from the biological insights that need further investigation.
Reviewer #2 (Public review):
Summary:
In this study, the authors aim to establish a calibrated framework for detecting RNA modifications using long-read sequencing and apply it to compare modification patterns between fibroblasts and neuron-like cells. The work combines long-read sequencing, in vitro transcribed controls, methyltransferase inhibition, and comparison to an orthogonal sequencing-based method in an attempt to derive filtering strategies that reduce false positive modification calls. The authors further apply this framework to explore differences in modification levels between the two cell types. The resulting dataset may be of interest to researchers working on RNA modification detection using long-read sequencing technologies. Independent datasets across additional cellular systems can be useful for benchmarking computational methods and evaluating the behavior of modification detection models. However, the conceptual advance of the analytical framework presented here remains somewhat unclear, as many aspects of the analysis closely resemble strategies that have already been described in recent benchmarking studies.
Strengths:
A clear strength of the study is the generation of a relatively large long-read sequencing dataset together with several useful experimental controls, including in vitro transcribed RNA and pharmacological inhibition of the methyltransferase enzyme responsible for installing this modification. These controls are helpful for illustrating the challenges associated with distinguishing high-confidence modification sites from background signals. The inclusion of two different human cellular systems also provides an additional dataset that may be useful for benchmarking and cross-validation in the field. The study addresses a practically relevant question for the community, namely, how to reduce false positive calls in long-read sequencing-based RNA modification analyses.
Weaknesses:
The main weakness of the manuscript is its limited methodological novelty. Much of the analytical framework presented here closely follows benchmarking strategies that have already been described in recent studies of RNA modification detection using long-read sequencing. Several previous studies have evaluated modification-aware basecalling approaches, discussed the need for stringent filtering strategies, and compared long-read sequencing-based predictions with orthogonal mapping approaches. The manuscript would therefore benefit from a deeper engagement with the recent benchmarking literature and a clearer explanation of what conceptual or methodological advance the present study provides beyond these earlier analyses.
We thank the reviewer for this constructive comment. At the time our study was conducted, relatively few benchmarking studies addressing these methodological aspects were available, and the analytical approach and thresholds applied in our study were therefore established independently of these reports. Several relevant studies (some initially as preprints) subsequently emerged within a relatively short period. Now that these studies are available, we have considered their findings, including the thresholds and filtering criteria applied, and have incorporated the relevant comparisons and methodological considerations into the revised manuscript. We have rewritten the introduction section (lines 99-117) and the discussion (lines 537-546, 547-549, 553-562) to incorporate newer studies and correctly integrate our study into the field. Furthermore, we consider these findings not to be supportive of our work, as they strengthen the overall interpretations and guidelines for ONT DRS users.
A second concern relates to the filtering strategy that forms the core of the proposed workflow. The manuscript applies several thresholds, including modification probability, stoichiometry, and read coverage cutoffs, but it is not clearly explained how these thresholds were determined. It remains unclear whether these cutoffs were derived from statistical calibration, empirical optimization using the presented dataset, or adopted from previous studies. Because the downstream conclusions depend strongly on these filtering choices, a clearer methodological justification would strengthen the work and help readers assess the robustness of the proposed framework.
We thank the reviewer for raising this important point. For the modification probability thresholds, we compared the distribution of modification probabilities in the biological samples with that observed in the IVT controls. In the revised manuscript, we have integrated a new statistical framework that enables an unbiased determination of modification probability cutoffs for each dataset (see lines 174184, Methods, as well as Fig 2 and Supplementary Figs 5, 8, 9 and 10). We further report error rates relative to IVT across different modification probability cutoffs and probability ranges (lines 186-200). The validread coverage threshold was introduced as an additional safeguard against sites supported predominantly by reads that were neither confidently called as the canonical nucleotide nor confidently assigned as modified (since valid coverage includes only reads supporting the canonical base or the corresponding modification above the selected probability threshold (≥0.98 or ≥0.99, depending on the modification) (lines 233-240). Finally, the stoichiometry cutoff was also informed by comparison with the IVT controls. Although stringent modification-probability filtering substantially reduced background, the majority of residual sites detected in the IVT controls occurred at low modification proportions, particularly below 10%. We therefore introduced the stoichiometry threshold to further distinguish biological modification signal from low-level background calls observed in the unmodified IVT reference (lines 241-249).
The interpretation of the comparison between the two modification detection approaches also appears somewhat overstated. Differences between the methods are frequently interpreted as evidence that one approach produces large numbers of false positive calls, but the analyses presented do not fully exclude alternative explanations such as differences in sensitivity, sequencing depth, or methodological biases. A more cautious interpretation of these discrepancies would therefore be appropriate.
Some discussion points also appear speculative. In particular, certain interpretations propose mechanistic explanations without presenting analyses that would allow these possibilities to be distinguished. Such interpretations would benefit from either additional supporting analyses or more cautious phrasing.
We have revised the relevant sections accordingly to provide a more cautious interpretation of these findings (lines 583-592).
From a methodological perspective, the statistical robustness of the thresholds used throughout the analysis could also be discussed in more detail. Given the relatively modest read coverage cutoff applied in the study, low stoichiometry estimates may be strongly influenced by sampling noise, and fixed stoichiometry thresholds may therefore not correspond to a consistent level of confidence across sites. In addition, the manuscript relies heavily on fixed modification probability cutoffs to define high-confidence calls, but it does not discuss whether these scores are statistically calibrated or how they relate to expected error rates. Neural network outputs are often not well-calibrated probabilities, and interpreting these values as direct confidence estimates can therefore be problematic. Finally, modification detection models trained on known modification sites may capture sequence-context patterns present in the training data, meaning that motif enrichment or positional distributions along transcripts may partly reflect model biases rather than purely biological signals. A brief discussion of these limitations would help readers better interpret the robustness of the proposed filtering strategy and the downstream biological conclusions.
We thank the reviewer for raising these important methodological considerations. We acknowledge that modification proportion estimates, particularly at sites with relatively low coverage and low modification proportions, may be influenced by sampling variability and may not accurately reflect the true modification proportion.
In terms of our filtering strategy, we performed false-positive rate analyses across multiple modification probability thresholds using repeated subsampling of biological and modification-free IVT datasets. In addition, we performed an independent change-point analysis to identify where modification signals in biological samples diverged from the IVT background. These analyses, together with the reduction in high-probability m6A calls following METTL3 inhibition, provide complementary empirical support for the applied filtering criteria.
We also acknowledge that sequence-context features learned by modification-detection models may influence predicted motif distributions. We therefore performed additional analyses of non-DRACH predictions, identifying both technical sources of erroneous calls and non-canonical sites consistent with genuine biological modifications, including STM2457-sensitive sites.
Overall, while the dataset may be of interest to the community, the extent to which the study advances current methodological understanding beyond recent benchmarking efforts remains limited.
We thank the reviewer for their comments and critiques. The revised version of the manuscript has addressed these points (as detailed above), and we consider the fact that other recently published benchmarking studies come to broadly similar conclusions provide additional support to our work.
Notably, however, our study remains the only one to use the same RNA as input for distinct sequencing approaches (i.e. nanopore vs. GLORI-Seq).
Minor comments:
The discussion of the "DRACH" versus "all-context" outputs would benefit from greater technical precision. The statement that the number of sites within DRACH motifs identified by the all-context approach was nearly identical to the number reported by the DRACH model may suggest that these outputs derive from fundamentally different predictive models. As I understand it, the underlying neural network is the same, whereas the distinction lies primarily in the classification context. Clarifying this explicitly in the manuscript would improve interpretability and avoid potential confusion for readers.
We have updated the text accordingly to ensure greater technical precision (lines 342-350).
The manuscript compares results obtained with different basecalling and modification settings but refers primarily to Dorado software versions. This may be misleading, as software version and model version are not necessarily equivalent. Different basecalling or modification models can be used with the same software release, and newer software versions may still use older models. For clarity and reproducibility, the authors should report the exact basecalling and modification model names used in the analyses rather than referring only to the Dorado software version.
We thank the reviewer for highlighting this distinction. We have revised the manuscript and figures accordingly and clarified the specific Dorado software version as well as the individual models used for each analysis (lines 315-330).
Reviewer #2 (Recommendations for the authors):
(1) To strengthen the methodological robustness of the proposed framework, the authors should clarify how the filtering thresholds used throughout the analysis were determined. If these thresholds were derived empirically from the present dataset, the authors could consider providing additional analyses demonstrating their robustness (for example, by showing how key results change across different probability or stoichiometry cutoffs). If the thresholds were adopted from previous studies, the relevant sources and the rationale for using them should be explicitly cited.
We thank the reviewer for this recommendation. As outlined before, the filtering thresholds used in the analysis were empirically informed by comparison of the biological samples with the unmodified IVT controls. In the revised manuscript, we have implemented a statistical framework that enables the identification of appropriate modification probability cutoffs on a per-sample basis. Specifically, we report error rates relative to the IVT controls across different modification probability cutoffs and probability ranges (lines 186-200), together with statistically supported identification of modification probability cutoffs for each biological sample (by comparing with the corresponding IVT reference, lines 220-230). These analyses support the modification probability thresholds applied in the downstream analyses.
We have also clarified the rationale for the additional filtering criteria, such as valid coverage and stoichiometry cutoffs. (lines 233-240 and 241-249).
(2) Because the analysis relies strongly on modification probability scores, the authors may also wish to discuss how these scores relate to expected error rates. If possible, providing an analysis of how false positive rates behave across probability thresholds using the IVT controls included in the study could help readers better interpret the meaning of the chosen cutoff values.
We thank the reviewer for this suggestion. As described above, we have now directly assessed how false-positive rates vary across modification probability thresholds using the unmodified IVT controls. In the revised manuscript, we report error rates relative to the IVT controls across a range of modification probability cutoffs and within discrete probability intervals (lines 203–213). These analyses demonstrate that false-positive rates decrease with increasing modification probability and provide an empirical basis for interpreting the selected probability thresholds.
(3) For the comparison between the two modification detection approaches, the authors may consider providing additional analyses that clarify the source of discrepancies between the methods. For example, examining whether differences depend on read coverage, sequence context, or site stoichiometry could help readers better understand the conditions under which the two approaches converge or diverge.
Thank you for this suggestion. We agree that analyses of read coverage, sequence context, and site stoichiometry could help explain the differences between the two methods. However, our current results suggest that the main reason for the discrepancy is the limited conversion efficiency of the GLORI dataset (see Fig. 3D). To properly assess the effects of coverage or sequence context, we would need a higher-quality GLORI dataset with better conversion efficiency. Generating such a dataset and repeating the analysis was beyond the scope of this study. We have therefore revised the text to make this limitation clearer and adjusted our interpretation accordingly (lines 343-361 and 583-593).
(4) The interpretation of modification patterns between fibroblasts and neuron-like cells could also be strengthened by additional analyses that account for potential confounding factors such as differences in transcript abundance, transcript coverage, or gene expression levels between the two datasets.
We thank the reviewer for this important suggestion. Our comparison was restricted to m6A sites shared between both cell types, minimizing potential biases from cell type-specific transcript detection. We agree that transcript abundance, coverage, and the effects of m6A on transcript stability and translation should be considered to assess the biological significance of these differences, but such analyses are beyond the scope of this study. We have therefore rephrased our statements more cautiously (lines 478-482).
(5) In the description of the modification detection models, the manuscript would benefit from clearer reporting of the exact basecalling and modification model names used in the analyses. Because software version and model version are not necessarily equivalent, specifying the model identifiers would improve clarity and reproducibility.
As mentioned above, we have revised the manuscript and figures accordingly and clarified the specific basecalling and modification models used for each analysis (lines 315-330).
(6) The manuscript would also benefit from clarifying the relationship between the "DRACH" and "allcontext" outputs, particularly whether these represent distinct predictive models or the same model applied in different classification contexts.
As mentioned above, we have revised the corresponding paragraph to avoid ambiguity regarding the models used in each analysis (lines 315-330).
(7) Finally, the manuscript would strongly benefit from a more detailed discussion of the relationship between the present analysis and previous benchmarking studies of long-read RNA modification detection, which would help readers understand how the current work relates to and extends existing efforts in the field.
We thank the reviewer for this helpful suggestion. We have expanded the discussion to better place our findings in the context of new benchmarking studies (538-547, 554-563).
Reviewer #3 (Public review):
In this study, the authors aim to establish a calibrated framework for identifying RNA chemical marks from direct RNA sequencing data using a modification-aware basecalling workflow, with a particular focus on N6-methyladenosine. By combining native RNA sequencing with an unmodified control transcriptome, enzyme inhibition, comparison across multiple software versions, and orthogonal validation using an independent mapping approach, the authors seek to define a best-practice pipeline for reducing falsepositive calls and improving confidence in quantitative interpretation across cell types.
A major strength of the work is the rigor of the benchmarking strategy. In particular, the inclusion of an unmodified control transcriptome is both important and useful, and the study provides compelling evidence that this control remains necessary for robust interpretation, despite being omitted in many current workflows. The comparison across software versions and the matched analysis with an independent sequencing-based approach also substantially strengthen the evidence presented. The work therefore makes a valuable contribution to the community by offering a more stringent analytical framework that will likely be broadly useful to groups applying native RNA sequencing to study RNA chemical marks.
The evidence supporting the main conclusions is solid overall. The authors convincingly show that stringent filtering substantially reduces false-positive calls and improves agreement with orthogonal approaches, particularly at highly modified sites. The observation that many sites are conserved across cell types, while showing differences in relative modification levels, is also supported by the presented analyses.
At the same time, several conceptual issues limit the strength of some downstream interpretations. Most importantly, the manuscript repeatedly refers to the reported values as "stoichiometry," whereas the underlying software output is more appropriately interpreted as a statistical estimate of the proportion of aligned reads classified as modified. This distinction is important because the conclusions regarding celltype differences rely on quantitative comparisons of these values. In addition, the current calling framework depends on successful canonical base assignment before modification calling, which raises an important limitation: sites with the strongest signal deviations may be underrepresented if they are more likely to be miscalled during basecalling. This issue may be especially relevant for RNA marks that induce stronger mismatch signatures than N6-methyladenosine and should be more explicitly discussed. Overall, the authors largely achieve their primary aim of establishing a more rigorous and broadly applicable analytical framework for direct RNA sequencing-based modification detection. The work is likely to have a meaningful impact on the field, particularly by reinforcing the importance of appropriate negative controls and benchmarking standards. With clearer framing of the quantitative outputs and explicit discussion of current software limitations, this study will serve as a highly useful resource for the community.
Reviewer #3 (Recommendations for the authors):
I have several conceptual and methodological concerns.
(1) Please avoid the term "stoichiometry". A major concern is the repeated use of the term "stoichiometry." In the context of Dorado-based nanopore calling, this is fundamentally a statistical estimate of the proportion of aligned reads classified as modified, rather than a direct biochemical measurement of true molecular stoichiometry or occupancy. I strongly recommend replacing this term throughout the manuscript with:
- predicted m6A/A proportion
- relative modification proportionThis distinction is especially important because the manuscript draws biological conclusions from differences in these values across cell types. At present, the language risks overstating the quantitative certainty of the method.
The reviewer is correct, and where applicable (text, figures), we have replaced the word “stoichiometry” with “predicted m6A/A proportion”.
(2) Canonical basecalling dependency requires explicit discussion.
A key conceptual concern is that the modification caller requires the base to first be correctly called as A before assigning it as m6A. This point needs to be explicitly addressed in the manuscript. Because the proportion is being calculated only from reads that are first aligned and called as canonical A, I strongly encourage the authors to include a mismatch analysis and explicitly frame the reported values as proportions derived only from correctly aligned A-containing reads. This raises an important question: are the strongest signal-distorting sites being systematically excluded from analysis because they are more likely to be basecalled incorrectly? This is an important limitation of the ONT Dorado/Modkit framework and should be discussed directly.
We thank the reviewer for raising this important point. We agree that the reported modification proportions do not capture all reads overlapping a given position. The proportions are calculated from reads in which the corresponding base is basecalled as an A and is confidently assigned as either canonical A or m6A above the selected modification-probability threshold. Reads in which the position is basecalled as a different nucleotide, contain a deletion, are assigned another modification, or have an m6A probability below the selected threshold are therefore not included in the proportion estimate. For the DRACH context model, reads containing a non-matching base at other positions within the required motif context are likewise excluded from modification calling. We have now clarified this limitation explicitly in the revised manuscript (lines 285-296). In addition, to assess the potential magnitude of this effect, we performed a mismatch analysis examining the percentage of reads in which the reference A was basecalled as another nucleotide and evaluated its relationship with the predicted m6A/A proportion across sites in our filtered dataset (lines 297-313). This analysis provides information that is not captured by the predicted modification proportions.
(3) The concern above (2) is likely stronger for Pseudouridine and Inosine RNA editing.
I think this issue is likely even more important for Pseudouridine and Inosine RNA editing than for m6A. For these modifications, the basecalling error at the modified site is expected to be substantially stronger than for m6A, which may leave very few aligned reads available for downstream proportion calculations. This could potentially explain why the authors require a higher modification probability threshold for these marks. I would strongly encourage the authors to explicitly discuss whether this basecalling dependency contributes to the need for the {greater than or equal to}0.99 threshold and whether this introduces a systematic detection bias.
To investigate the potential impact of basecalling errors on modification proportion estimates, we examined alternative nucleotide calls at modification sites, including pseudouridine and inosine sites, focusing on positions where non-reference basecalls accounted for ≥25% of the reads overlapping the site (lines 297-305). At pseudouridine sites, the predominant alternative basecall was C, whereas at inosine sites, G was most frequently observed. These results indicate that modification-associated basecalling errors may lead to the exclusion of a fraction of reads from downstream modification proportion calculations. Consequently, the reported modification proportions may be underestimated at sites where the presence of a modification strongly affects canonical basecalling, introducing a potential detection bias.
Importantly, however, the modification probability thresholds used in our analysis were derived empirically by comparing modification probability distributions in the biological samples with those observed in the corresponding IVT and STM2457-treated controls. Basecalling accuracy was not considered in establishing these thresholds.
(4) Need better framing of multiple modifications on the same transcript.
In the Introduction, I would encourage the authors to better position nanopore DRS relative to short-read approaches by explicitly noting that short-read methods are also limited in their ability to resolve multiple modifications on the same transcript molecule. This is an important conceptual advantage of native longread sequencing and deserves clearer emphasis early in the manuscript.
We agree that the ability to examine the co-occurrence of multiple modifications on individual RNA molecules represents an important distinction between nanopore DRS and short-read approaches. We have therefore revised the Introduction to explicitly note the limitation (lines 79-80, 83-84).
(5) Clarify unclear experimental step.
The stoichiometry distribution experiment (lines 189-195) in the Results section is difficult to follow.
We thank the reviewer for pointing this out. We have revised and restructured this section to more clearly describe the stoichiometry analysis (lines 255-274).
(6) Discussion of non-DRACH sites.
The manuscript interprets many non-DRACH GLORI sites as likely false positives. While this is a reasonable interpretation, I would like the authors to also discuss whether true cellular m6A may exist at non-DRACH motifs, and whether any non-canonical sites identified by nanopore DRS may represent biologically real events rather than technical artifacts.
We agree and have added a new analysis (Fig. S23 and S24) addressing this point. Most non-DRACH calls derive from technical causes (SNPs, homopolymers, proximity to true DRACH sites), but a subset conforming to DRAC motifs, e.g., are STM2457-sensitive, indicating they likely represent genuine METTL3-dependent m6A sites. We have added a description of the new figures in the results section 425456) and discuss this further in lines 564-579.
This is especially important given emerging evidence that non-canonical motifs can occasionally occur in vivo.
We thank the reviewer and have added corresponding data and discussion, as pointed out above. While most all-context non-DRACH calls reflect similar technical artifacts, we identify STM2457-sensitive GGAUU sites, e.g., and a distinct STM2457-insensitive, METTL16-like class (e.g., ACAGA in MAT2A), supporting that some non-canonical calls represent true m6A sites.
Regarding lines 236-240: Are there also sites from nanopore sequencing that are unexpectedly positive?
We thank the reviewer for this question and refer to our newly added figures (Figure S23 and S24) and explanations described above.
(7) Strongest point of the paper. I want to emphasize that one of the strongest contributions of this study is its clear demonstration that the modification-free transcriptome control is necessary. This should be stated even more explicitly as a best-practice recommendation, as the field increasingly trends toward analyses that omit such controls.
We thank the reviewer for highlighting this important aspect of our study. We have strengthened this point in the Discussion (lines 610-611) and now more explicitly recommend using a modification-free transcriptome control as a best practice for reliable RNA modification detection by DRS.
Overall, I find this to be a strong manuscript, with its methodological rigor being a major strength. Addressing the concerns above would substantially improve both the clarity and interpretability of the work.
We thank the reviewer for their positive assessment and constructive feedback. We believe that the revisions have improved the clarity of the manuscript and the interpretability of our findings.
-
-
-
-
Link to paper: elifesciences.org/reviewed-pre... If you want to discuss pipelines that survive reviewer scrutiny, contact us at Trinobia: www.trinobia.com/contact/ @deniseohnezeit.bsky.social @anguswilson.bsky.social
-
They benchmark multiple Dorado versions using controls many workflows still skip. In vitro transcribed RNA, METTL3 inhibition, and side by side GLORI on the same RNA. This is the baseline many studies lack. #RNAmodifications #m6Aseq #PeerReview #GenomicsResearch #BioinformaticsTools
-
eLife Assessment
This valuable study compares orthogonal approaches for detecting RNA chemical modifications and provides a helpful framework for improving the reliability of direct RNA sequencing-based identification of RNA modifications. The evidence supporting the technical benchmarking claims is solid. However, support for the broader biological conclusions is not as strong, and the quantitative interpretation of the results, as well as the limitations of the underlying models, would benefit from further clarification.
-
Reviewer #1 (Public review):
Summary:
The authors set out to evaluate how accurately direct sequencing of RNA can identify and quantify several chemical modifications on RNA molecules, focusing primarily on m6A. A central goal of the work is to compare this approach with an independent chemical-based method (glyoxal and nitrite-mediated deamination of unmethylated adenosines (GLORI), using the same RNA samples, in order to assess reproducibility, false-positive signals, and sensitivity across a range of detection strategies. The authors further aim to demonstrate the biological utility of this approach by applying it to two human cell types, primary human fibroblasts and HD10.6 neurons. While the manuscript also reports detection of additional RNA modifications (pseudouridine and m5C, the depth of analysis and strength of controls are …
Reviewer #1 (Public review):
Summary:
The authors set out to evaluate how accurately direct sequencing of RNA can identify and quantify several chemical modifications on RNA molecules, focusing primarily on m6A. A central goal of the work is to compare this approach with an independent chemical-based method (glyoxal and nitrite-mediated deamination of unmethylated adenosines (GLORI), using the same RNA samples, in order to assess reproducibility, false-positive signals, and sensitivity across a range of detection strategies. The authors further aim to demonstrate the biological utility of this approach by applying it to two human cell types, primary human fibroblasts and HD10.6 neurons. While the manuscript also reports detection of additional RNA modifications (pseudouridine and m5C, the depth of analysis and strength of controls are greatest for m6A, which forms the primary focus of the study
Strengths:
A strength of this work is the direct comparison of two distinct measurement approaches performed on the same RNA input material; this has not been done in other recently published benchmarking studies evaluating the utility of the recent direct RNA sequencing for calling m6A. The authors systematically test multiple analysis models and show that, when appropriate filtering is applied, detection of modified sites is reproducible across software versions. The use of synthetic RNA standards and METTL3 inhibitors as negative controls helps to reinforce the overall results.
The data show good agreement between the two methods at higher m6A modification levels, supporting the conclusion that direct RNA sequencing can reliably detect high-confidence modification sites. The authors also demonstrate that this approach can, in principle, provide information at the level of individual RNA variants (although only one example was provided), which is difficult to achieve with short-read methods. The methodology described here is likely to be useful to others seeking to apply similar approaches to identify and quantify m6A. The study also explores the detection of other RNA modifications, which highlights the broader potential of the approach, although these analyses are necessarily more exploratory given the more limited controls and data available.
Weaknesses:
Despite these strengths, several issues limit the interpretation of the results and should be clarified for readers.
First, the authors appropriately address false-positive signals by estimating expected false-positive rates and by quantitatively comparing sequence motif enrichment before and after filtering. These analyses provide important support for the use of stoichiometry-based thresholds and demonstrate that filtering substantially improves specificity. However, even after filtering, a subset of detected sites remains outside the expected sequence context. It therefore remains unclear to what extent these non-canonical sites reflect genuine biology versus residual technical artifacts.
Second, claims regarding the ability of direct RNA sequencing to resolve modification patterns across different RNA variants are supported by very limited evidence. The conclusion that this approach provides superior isoform-level quantification relative to short-read methods is based largely on a single gene example. While this case is interesting, it does not establish how widespread or general this advantage is. A broader analysis indicating how many genes show isoform-specific modification patterns detectable by this method, and how often these are missed by the comparison approach, would be necessary to support a general claim.
Third, the biological interpretation of cell type-specific differences in modification levels remains underdeveloped. Although differences in modification stoichiometry are reported between fibroblasts and neuron-derived cells, the functional consequences of these differences are not addressed. It is unclear whether changes in modification levels are associated with differences in RNA abundance, stability, or translation. As a result, statements suggesting that these modifications fine-tune core cellular pathways are speculative and should either be supported with additional analyses or framed more cautiously.
Related to this point, differences in gene expression between the two cell types are a potential confounding factor. The pathway enrichment patterns presented appear biased toward particular functional categories, but without clear control for differential gene expression, it is difficult to determine whether the observed enrichment reflects cell type-specific regulation of RNA modification or simply differences in which genes are expressed. Clarifying how background gene sets were defined for these analyses would help readers interpret the results.
The manuscript also suggests broader differences in overall modification levels between cell types, but this is not validated using an independent global assay. An orthogonal measurement of total modification levels on polyadenylated RNA (for example, dot blot) would help place site-specific stoichiometry differences in a clearer biological context.
Finally, the effects of the METTL3 inhibitor on these cell types are not fully characterized. While changes in m6A modification patterns are reported following treatment, the manuscript does not address whether the treatment affects cell growth or viability.
Appraisal of conclusions and impact:
Overall, the study provides an informative technical assessment of direct RNA sequencing for modification detection and establishes clear conditions under which the method performs well. The evidence strongly supports conclusions related to technical benchmarking, reproducibility, and the importance of filtering and controls, particularly for m6A. In contrast, conclusions regarding isoform-specific regulation and cell type-specific biological roles of RNA modification are less well supported by the data currently presented, and would benefit from either additional analysis or more restrained interpretation.
The work is likely to have a meaningful impact as a practical reference for researchers using direct RNA sequencing, particularly by clarifying sources of false positives and the value of appropriate controls. With clearer limits placed on biological interpretation or more data presented in support of the biological interpretation, the study would serve as a valuable reference for users seeking to apply these technologies reliably.
-
Reviewer #2 (Public review):
Summary:
In this study, the authors aim to establish a calibrated framework for detecting RNA modifications using long-read sequencing and apply it to compare modification patterns between fibroblasts and neuron-like cells. The work combines long-read sequencing, in vitro transcribed controls, methyltransferase inhibition, and comparison to an orthogonal sequencing-based method in an attempt to derive filtering strategies that reduce false positive modification calls. The authors further apply this framework to explore differences in modification levels between the two cell types.
The resulting dataset may be of interest to researchers working on RNA modification detection using long-read sequencing technologies. Independent datasets across additional cellular systems can be useful for benchmarking …
Reviewer #2 (Public review):
Summary:
In this study, the authors aim to establish a calibrated framework for detecting RNA modifications using long-read sequencing and apply it to compare modification patterns between fibroblasts and neuron-like cells. The work combines long-read sequencing, in vitro transcribed controls, methyltransferase inhibition, and comparison to an orthogonal sequencing-based method in an attempt to derive filtering strategies that reduce false positive modification calls. The authors further apply this framework to explore differences in modification levels between the two cell types.
The resulting dataset may be of interest to researchers working on RNA modification detection using long-read sequencing technologies. Independent datasets across additional cellular systems can be useful for benchmarking computational methods and evaluating the behavior of modification detection models. However, the conceptual advance of the analytical framework presented here remains somewhat unclear, as many aspects of the analysis closely resemble strategies that have already been described in recent benchmarking studies.
Strengths:
A clear strength of the study is the generation of a relatively large long-read sequencing dataset together with several useful experimental controls, including in vitro transcribed RNA and pharmacological inhibition of the methyltransferase enzyme responsible for installing this modification. These controls are helpful for illustrating the challenges associated with distinguishing high-confidence modification sites from background signals. The inclusion of two different human cellular systems also provides an additional dataset that may be useful for benchmarking and cross-validation in the field. The study addresses a practically relevant question for the community, namely, how to reduce false positive calls in long-read sequencing-based RNA modification analyses.
Weaknesses:
The main weakness of the manuscript is its limited methodological novelty. Much of the analytical framework presented here closely follows benchmarking strategies that have already been described in recent studies of RNA modification detection using long-read sequencing. Several previous studies have evaluated modification-aware basecalling approaches, discussed the need for stringent filtering strategies, and compared long-read sequencing-based predictions with orthogonal mapping approaches. The manuscript would therefore benefit from a deeper engagement with the recent benchmarking literature and a clearer explanation of what conceptual or methodological advance the present study provides beyond these earlier analyses.
A second concern relates to the filtering strategy that forms the core of the proposed workflow. The manuscript applies several thresholds, including modification probability, stoichiometry, and read coverage cutoffs, but it is not clearly explained how these thresholds were determined. It remains unclear whether these cutoffs were derived from statistical calibration, empirical optimization using the presented dataset, or adopted from previous studies. Because the downstream conclusions depend strongly on these filtering choices, a clearer methodological justification would strengthen the work and help readers assess the robustness of the proposed framework.
The interpretation of the comparison between the two modification detection approaches also appears somewhat overstated. Differences between the methods are frequently interpreted as evidence that one approach produces large numbers of false positive calls, but the analyses presented do not fully exclude alternative explanations such as differences in sensitivity, sequencing depth, or methodological biases. A more cautious interpretation of these discrepancies would therefore be appropriate.
Some discussion points also appear speculative. In particular, certain interpretations propose mechanistic explanations without presenting analyses that would allow these possibilities to be distinguished. Such interpretations would benefit from either additional supporting analyses or more cautious phrasing.
From a methodological perspective, the statistical robustness of the thresholds used throughout the analysis could also be discussed in more detail. Given the relatively modest read coverage cutoff applied in the study, low stoichiometry estimates may be strongly influenced by sampling noise, and fixed stoichiometry thresholds may therefore not correspond to a consistent level of confidence across sites. In addition, the manuscript relies heavily on fixed modification probability cutoffs to define high-confidence calls, but it does not discuss whether these scores are statistically calibrated or how they relate to expected error rates. Neural network outputs are often not well-calibrated probabilities, and interpreting these values as direct confidence estimates can therefore be problematic. Finally, modification detection models trained on known modification sites may capture sequence-context patterns present in the training data, meaning that motif enrichment or positional distributions along transcripts may partly reflect model biases rather than purely biological signals. A brief discussion of these limitations would help readers better interpret the robustness of the proposed filtering strategy and the downstream biological conclusions.
Overall, while the dataset may be of interest to the community, the extent to which the study advances current methodological understanding beyond recent benchmarking efforts remains limited.
Minor comments:
The discussion of the "DRACH" versus "all-context" outputs would benefit from greater technical precision. The statement that the number of sites within DRACH motifs identified by the all-context approach was nearly identical to the number reported by the DRACH model may suggest that these outputs derive from fundamentally different predictive models. As I understand it, the underlying neural network is the same, whereas the distinction lies primarily in the classification context. Clarifying this explicitly in the manuscript would improve interpretability and avoid potential confusion for readers.
The manuscript compares results obtained with different basecalling and modification settings but refers primarily to Dorado software versions. This may be misleading, as software version and model version are not necessarily equivalent. Different basecalling or modification models can be used with the same software release, and newer software versions may still use older models. For clarity and reproducibility, the authors should report the exact basecalling and modification model names used in the analyses rather than referring only to the Dorado software version.
-
Reviewer #3 (Public review):
In this study, the authors aim to establish a calibrated framework for identifying RNA chemical marks from direct RNA sequencing data using a modification-aware basecalling workflow, with a particular focus on N6-methyladenosine. By combining native RNA sequencing with an unmodified control transcriptome, enzyme inhibition, comparison across multiple software versions, and orthogonal validation using an independent mapping approach, the authors seek to define a best-practice pipeline for reducing false-positive calls and improving confidence in quantitative interpretation across cell types.
A major strength of the work is the rigor of the benchmarking strategy. In particular, the inclusion of an unmodified control transcriptome is both important and useful, and the study provides compelling evidence that …
Reviewer #3 (Public review):
In this study, the authors aim to establish a calibrated framework for identifying RNA chemical marks from direct RNA sequencing data using a modification-aware basecalling workflow, with a particular focus on N6-methyladenosine. By combining native RNA sequencing with an unmodified control transcriptome, enzyme inhibition, comparison across multiple software versions, and orthogonal validation using an independent mapping approach, the authors seek to define a best-practice pipeline for reducing false-positive calls and improving confidence in quantitative interpretation across cell types.
A major strength of the work is the rigor of the benchmarking strategy. In particular, the inclusion of an unmodified control transcriptome is both important and useful, and the study provides compelling evidence that this control remains necessary for robust interpretation, despite being omitted in many current workflows. The comparison across software versions and the matched analysis with an independent sequencing-based approach also substantially strengthen the evidence presented. The work therefore makes a valuable contribution to the community by offering a more stringent analytical framework that will likely be broadly useful to groups applying native RNA sequencing to study RNA chemical marks.
The evidence supporting the main conclusions is solid overall. The authors convincingly show that stringent filtering substantially reduces false-positive calls and improves agreement with orthogonal approaches, particularly at highly modified sites. The observation that many sites are conserved across cell types, while showing differences in relative modification levels, is also supported by the presented analyses.
At the same time, several conceptual issues limit the strength of some downstream interpretations. Most importantly, the manuscript repeatedly refers to the reported values as "stoichiometry," whereas the underlying software output is more appropriately interpreted as a statistical estimate of the proportion of aligned reads classified as modified. This distinction is important because the conclusions regarding cell-type differences rely on quantitative comparisons of these values. In addition, the current calling framework depends on successful canonical base assignment before modification calling, which raises an important limitation: sites with the strongest signal deviations may be underrepresented if they are more likely to be miscalled during basecalling. This issue may be especially relevant for RNA marks that induce stronger mismatch signatures than N6-methyladenosine and should be more explicitly discussed.
Overall, the authors largely achieve their primary aim of establishing a more rigorous and broadly applicable analytical framework for direct RNA sequencing-based modification detection. The work is likely to have a meaningful impact on the field, particularly by reinforcing the importance of appropriate negative controls and benchmarking standards. With clearer framing of the quantitative outputs and explicit discussion of current software limitations, this study will serve as a highly useful resource for the community.
-
-
-
-
New preprint benchmarks #Nanopore direct RNA sequencing for quantitative m6A mapping, revealing best practice filters and cell type specific #epitranscriptomic differences in fibroblasts and neurons. #RNAseq 📄 https://doi.org/10.1101/2025.11.02.686099 👤 EVBC member: Daniel Depledge
-
Checkout our new preprint (www.biorxiv.org/content/10.1...) where we optimize Dorado on RNA from multiple cell types and directly compare to GLORI-seq.
-
There's a lot of excitement about Dorado, the new modification-aware basecaller from Oxford Nanopore Technologies@nanoporetech.com but the accuracy and best practices are not widely understood.
-
Establishing benchmarks for quantitative mapping of m6A by Nanopore Direct RNA Sequencing. #DirectRNAseq #RNAsequencing #RNAmodifications #m6A #Genomics #Bioinformatics @nanoporetech.com 🧪🧬 🖥️ www.biorxiv.org/content/10.1...
-
Establishing benchmarks for quantitative mapping of m6A by Nanopore Direct RNA Sequencing. #DirectRNAseq #Sequencing #RNAmodifications @nanoporetech.com www.biorxiv.org/content/10.1...
-
-
Establishing benchmarks for quantitative mapping of m6A by Nanopore Direct RNA Sequencing doi.org/10.1101/2025...
-
Establishing benchmarks for quantitative mapping of m6A by Nanopore Direct RNA Sequencing https://www.biorxiv.org/content/10.1101/2025.11.02.686099v1
-
Establishing benchmarks for quantitative mapping of m6A by Nanopore Direct RNA Sequencing https://www.biorxiv.org/content/10.1101/2025.11.02.686099v1
-