We thank Dr. Niranjan Nagarajan for reviewing the manuscript and providing his comments. Following his suggestions, we have revised the manuscript and have responded to all his queries below. We believe this has improved the manuscript.

Reviewer’s comments:

RNAtor looks like a simple and easy to use application. However this could also be its drawback in that it may be too simplistic. The authors should show some evidence that the parts they omit do not significantly impact RNAtor's guidance to users.

Authors’ response:

We agree with the reviewer that the RNAtor v1.0 is simplistic and some of the assumptions can affect the recommendations. However, the tool is intended to be used by biologists who use RNA-seq as a replacement to existing technology like microarray and qPCR, to ask simpler biological questions. Many small biology labs, we believe, are still in this group. The tool is not intended to serve advanced users. We are well aware of its limitations and intend to expand the scope of the tool in its future versions, by introducing biases that mimick various experimental conditions into the simulation phase. Nevertheless, we believe that the validation of the recommendations resulting from training on simulated RNA-seq data that has not yet incorporated various biological biases, with real data from Saccharomyces cerevisiae provides strong evidence that our assumptions do not significantly impact RNAtor's guidance to users.

Abstract:

Reviewer’s comments:

1) Perhaps the first sentence of the abstract should be reworded as it is a bit odd to say that "RNA Sequencing ... understands transcript structures" etc.

Authors’ response:

We thank the reviewer to point this. We have re-worded this sentence in the revised manuscript.

Reviewer’s comments:

2) Its unnecessary to mention the "number of lanes" when the "number of reads" has been highlighted.

Authors’ response:

Following the reviewer’s suggestion, we have removed the use of the term in the revised manuscript.

Reviewer’s comments:

3) It is s not clear why a mobile application is necessary. Is a web-based tool like EDDA not sufficient?

Authors’ response:

We agree with the reviewer that there are web-based tools, like EDDA and Scotty, which we have now elaborated in the revised manuscript, that can do similar job. However, we strongly believe that a  mobile application offers a lot more flexibility, ease of navigation, and user-friendliness, compared to a web-based tool. Besides, many features of the App are available for offline use and the user has the flexibility to use it on the work bench or anywhere convenient. Additionally, Apps use a different framework to run code and hence, run faster than mobile websites that typically use javascript to run their code, and store the data locally on the mobile device, unlike websites that store data on web servers.

Introduction:

Reviewer’s comments:

4) The last sentence is not very clear in terms of what is being done here.

Authors’ response:

Following the reviewer’s suggestion, we have modified the sentence in the revised manuscript.

Reviewer’s comments:

5) As it is typical to present and discuss prior work in the introduction, I believe it is important to add this component. Some of this could perhaps come from the current discussion section.

Authors’ response:

We have now modified the introduction to include references on tools aiding RNA-seq experimental design.

Implementation:

Reviewer’s comments:

6) What were the parameters used for running Polyester? In particular, different experimental conditions will likely have different expression profiles, biological variability dictating the dispersion parameters and perhaps other factors like splicing complexity, gene families etc. that all affect RNA-seq analysis. How does RNAtor account for these? How were the various differential analysis softwares run? They usually have cutoffs that can be chosen to make their results more or less stringent.

Authors’ response:

The per_reads_transcript, num_reps, fold_changes, in addition to the input ‘fasta’ and output ‘outdir’ parameters were exercised in polyester simulations. All differential expression analysis softwares were run with default cut-offs.

We have now mentioned these in the revised manuscript

Results:

Reviewer’s comments:

7) The detection of differentially abundant RNAs using RNA-seq is impacted by the relative abundance of the transcript. How is this taken into account in table 1?

Authors’ response:

In the current version of the tool implementation, the simulated data does not take into account the relative abundance of transcript, for e.g. in relation to a control gene or any other gene of interest. Hence, our recommendations (Table 1) are not affected by these. However, differences in relative abundances of true control genes between treatment and control would lead to over- or under-estimation of DEGs, and therefore, the base-line assumption of log ₂FC=0 should be accordingly adjusted.

We have acknowledged this limitation in the revised manuscript.

Reviewer’s comments:

8) In figure 2, are these all true positives? What about false positives? Usually there is a tradeoff and a user needs to be aware of this.

Authors’ response:

Figure 2 are all detected DEGs, i.e., both true and false positives.  Supplementary Figure S3 gives the true positive and false positive trends separately. Yes, there is a tradeoff between sensitivity (true postives) and specificity (absence of false positives).  Kallisto performs with a high sensitivity while compromising on specificity, while CuffDiff performs with a high specificity while compromising on sensitivity. We have elaborated this in the revised manuscript.

Reviewer’s comments:

9) Figure 3 does not have a legend.

Authors’ response:

We thank the reviewer to point this out and have now added a legend to Figure 3.

Reviewer’s comments:

10) It is not clear to me what transcript recovery is referring to and why that is relevant here.

Authors’ response:

Transcript recovery refers to the length of the transcript as assembled by Tophat, detected  as differentially expressed by EdgeR or CuffDiff or DESeq2, in relation to the actual length, as per simulations. It is possible to estimate this parameter only for these three tools, since they offer a handle to the actual transcript Ids. This has been clarified in the revised manuscript.

Reviewer’s comments:

11) Supplementary figure 4 needs better resolution and font sizes.

Authors’ response:

Supplementary Figure 4 with improved resolution has now been added.

Reviewer’s comments:

12) The claims made in the section "Detection specificity and transcript recovery by DE tools" are not obvious from the figures shown.

Authors’ response:

The results in this section have been elaborated further with reference to Supplementary Figure S4 in the revised manuscript. With reference to the claim made about detection specificity citing Supplementary Figure S3, we found that CuffDiff performs with high specificity at the loss of sensitivity, with the opposite being true for Kallisto-Sleuth.

We thank Dr. Daisuke Komura for his time and timely review. We have addressed all his queries below and have incorporated his suggestions in the revised version of the manuscript, which we believe is much improved now.

Response to Reviewers’ comments

Major:

Reviewer’s comments:

1) In general, some reads in RNA-seq analysis are removed due to their low sequencing quality or cannot be mapped probably due to library contamination of rRNA or other species’ RNA. Does this software consider the effect? If not, it could underestimate the number of required reads. 

Authors’ response:

No, the software does not consider this effect. The simulated reads correspond to error-free/-corrected reads. The add_error and add_platform_error features are added in a recent release of Polyester, not available when we used Polyester.

Reviewer’s comments:

2) How does RNAtor calculate the number of reads required for DEG detection? The authors claim that it is calculated based on the number of DEGs at its peak in both real and simulated data but the exact algorithm is not shown. Specifically, how were the peak values calculated and how were the peak values of real and simulated data merged? 

Authors’ response:

The detected DEGs by various tools were plotted as a function of simulated reads and replicates (Figure 2). The saturation point of the DEG curves, i.e. the peak was used to recommend a certain no of reads given the no of replicates, to detect optimal DEGs. Validation for this recommendation was obtained from extrapolation of the read numbers according to the Saccharomyces transcriptome size (Figure 3).

Minor: 

Reviewer’s comments:

3) Which does the term ‘replicates’ mean in this manuscript, technical replicate or biological replicate?

Authors’ response:

We thank the reviewer to point this out. The replicates used were technical. We have made this change in the revised manuscript.

Reviewer’s comments:

4) Implementation: “… workflow followed by differential expression analysis using five tools:…” I think four instead of five is correct because Kallisto does not use outputs of Tophat-Cufflinks pipeline (Supplementary Figure 1).

Authors’ response:

Kallisto is used twice; first, with the genome-guided paradigm and second, with de novo assembly using Trinity. In the first scenario, the Tophat-Cufflinks alignments (.bam) were converted to reads (.fastq) to be used with Kallisto along with the 3MB transcriptome as the reference. In the second scenario, the de novo assembled transcriptome was used as the reference along with the simulated reads with Kallisto. This has been clarified in the revised manuscript.

Reviewer’s comments:

5) Figure 2: How many DEGs were included in total in the simulated datasets? Sensitivity (%) is preferable to the number of DEGs in this case. The number of replicate is a discrete value but the slope in this figure is smooth. What do the numbers in fold change (1.2 0.83-5, 0.2)  mean? Figure 2: I think "fold change (1.2,0.83 - 5,0.2)" is confusing so it might be better to change it to another one such as ((>1.2 or <1/1.2) to (>5 or <1/5)).

Authors’ response:

The total no of simulated DEGs are 363. TP and FP curves are represented in Supplementary Figure S3. We have changed the representation for replicate numbers. The simulated fold changes were 1.2X to 5X, in both directions, i.e. 1.2,1/1.2=0.83 and 5,1/5=0.2. Fold change (1.2,0.83 - 5,0.2)" is changed to ((>1.2 or <1/1.2) to (>5 or <1/5)).

Reviewer’s comments:

6) Figure 3: Why were the result of 5x and 4x merged? What does each line indicate?  # reads = 0 means nothing thus should be removed. Figure 3: What do the lines indicate in each subplot?

Authors’ response:

The number of DEGs obtained separately at 4X and 5X were not sufficient to draw any conclusions, Hence they were merged as 4X+5X or >4X. We have removed the #reads=0 data point. We have updated the Figure 3 legend.