DRETools: A tool-suite for differential RNA editing detection

Recent tools to detect RNA editing have expanded our understanding of epitranscriptomics, linking changes in RNA editing to both disease and normal cellular processes. However, the research community currently lacks tools for determining if change in RNA editing or "differential editing" has occurred. To meet this need, we present DRETools, a command-line tool-set for finding differential editing among samples, editing islands, and editing sites.

Introduction RNA editing is a class of epitranscriptomic post-transcriptional modification found throughout metazoa consisting of the abundant conversion of adenosine-to-inosine (A-to-I) by ADARs (adenosine deaminases acting on RNA) and rare conversion of cytosine-to-uridine (C-to-U) by APOBEC (apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like) 1 . RNA editing is particularly interesting as it is detectable as A-to-G and C-to-T mismatches to the reference genome within standard RNA-sequencing data via specialized computational pipelines 2 . An increasing number of studies link changes in editing at specific sites or clusters-of-sites to diseases, such as epilepsy and atherosclerosis 3,4 . Yet, no software for detecting differential editing is available. To meet this need, we present DRETools 5 : 1) to calculate units that help reduce sample-bias, similar to FPKM for RNA expression; and 2) to find differentially edited sites and editing islands (i.e., clusters of editing sites) 6 . Further, we showcase two examples of finding differential editing and related tasks with DREtools 7

Implementation
DRETools can be run via command-line by typing "dretools", which will print the main help menu. The main help menu contains a list of operations that are available from dretools with short descriptions of each operation's purpose. To run an operation, type dretools followed by the operation name. Further detail on each operation, including available command-line arguments and usage examples, can be found by running an operation with the --help argument. On the main help menu, operations are organized into sub-headings based on similar functions. Further detail of each sub-heading and corresponding operations can be found in the following sections.

Units
One fundamental problem of between groups of samples is a lack of standardized units for describing editing within samples, editing islands, and sites. To this end, DRETools implements Editing Per Kilobase (EPK) based on "overall editing" (OE) 8 . EPK builds upon OE by considering both A-to-G and C-to-T transitions, excludes editing sites with 100% edited bases as potential mutations, and scaling by 10 3 for readability (similar to FPKM). EPK is calculated by dividing the total number of "edited" bases by the total number of bases overlapping known editing sites and multiplying by 10 3 . In addition to samples, DRETools can compute EPKs for editing islands and sites. Sample-wise editing can be computed with the "sample-epk" function and can be thought of as the global-editing-rate, whereas, the EPK of islands and sites can be computed with "region-epk" and "edsite-epk" respectively, and thought of as the "local-editing-intensity".

Differential editing
Recently, a method was developed to find differentially edited sites between epileptic or control mouse hippocampi 3 . However, methods capable of comparing different tissues are also needed. The problem is that unless the global-editing-rates are similar, we cannot determine if changes are due to differing global-editing-rates or other phenomena, such as competition with N6-methyladenosine (m6A) 9 . Furthermore, ADARs have been described to edit both specific sites in some cases and nonspecifically within small regions in other cases 10 . Therefore, in addition to individual editing sites, looking at the clusters of editing is also of interest. DRETools addresses both these issues by allowing the normalization of both the global-editing-rate and site or island local-editing-intensity in EPK and testing for differential editing using a linear model (LM) with the formula: "logFeatureEPK ~ logSampleEPK + featureLength + aver-ageReadDepth" (features can be sites or islands), which adjusts expectations for what constitutes differential editing.

Merge and stats
DRETools also includes various helper functions. For example, the merge section contains functions to find editing islands 6 and create consensus sets of editing sites by merging sites from multiple samples. Finally, the stats heading contains functions that calculate useful information about editing at the sample, gene, and site levels, such as the editable area or the number of editing sites falling in 3'/5'-untranslated regions, introns, or exons.
Operation A standard laptop computer with the latest version of R and Python3 will handle most applications.

Results
To illustrate the utility of DRETools, we surveyed differential editing in human umbilical vein endothelial cells (HUVEC) transfected with either a siRNA against ADAR1 or against a random sequence (control) 4 and the immortalized cell lines GM12787 and K562 11 . First we surveyed sample-wise editing using the function "sample-epk." (Figure 1A, B). Using EPK reduces variation within groups compared to the usage of number of editing sites. For example, the coefficient of variance drops from 0.21 to 0.05 for the silenced ADAR1 group and 0.52 to 0.01 for the control group. Similarly, when comparing the immortalized cell lines, the coefficient of variance is reduced from 0.57 to 0.25 and 0.46 to 0.11, respectively ( Figure 1C, D).
Next, we compared the EPKs of editing islands within the immortalized cell lines using "epk-region". Using EPK to represent editing islands as opposed to the number of edited bases reduces the coefficient of variance from 0.60 ± 0.21 to 0.31 ± 0.11 (p=2E-30). Finally, we tested for differential editing using the functions "region-diff" for islands and "site-diff" for editing sites ( Figure 1E-H). Comparing silenced ADAR1 to the control, the LM yielded a uniform distribution of p-values. In contrast, when using t-test applied to the same data, the distribution of p-values is shifted to the left and exhibits greater skew. However, in the immortalized cell lines, p-values calculated by the LM are more leftward skewed while p-values from the t-test became more uniformly distributed. This provides evidence that the LM can effectively reduce type I errors when testing for differential editing. For example, the LM correctly recognizes that most of the differences between the silenced ADAR1 and control groups arise from the reduction of the global-editing-rate in the silenced samples. Whereas the t-test, which does not consider the global-editing-rates, finds many differentially edited sites and islands. Conversely, when comparing the immortalized cell lines, despite the large difference in EPK, many differentially edited sites and islands are detected. While deeper biological validation is needed to be certain, these could be instances of some other phenomena, such as m6A 9 , affecting the editing in individual sites or islands.

Conclusions
DRETools is a command-line tool suite for finding differentially edited sites and islands. It allows users to calculate units that reduce sample-bias and find differentially edited sites and islands even when the global-editing-rate of groups being compared is different. Furthermore, it also includes a variety of other features for exploring RNA editing. These make DRETools a valuable tool for further investigating epitranscriptomics.

Data availability
All RNA-seq data are publically available and were downloaded from the NCBI SRA database 12 . The HUVEC data sets were generated by Stellos et al., 2016 4   The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
Other authors proposed similar indices to detect editing activity in a sample. For example, Paz-Yaacov introduced the Alu editing index, a robust measure that can be calculated also on other additional genomic properties (recoding sites, conserved sites and so on). This index has been successfully used in several cases and authors need to perform appropriate comparisons.
Regarding statistical tests used to detect differential editing, authors implement a linear model and the t-test. In figure 1 (panels E to H), p-values distributions are shown and great differences seem to appear. Authors should discuss the reason why of these observed discrepancies. I suggest authors to check the use of non parametric tests since they could be robust in case of small samples or when users cannot easily establish if normality and other assumptions are encountered.
Additionally, the tool does not take into account the correction for multiple testing. So it needs to be implemented.
In humans, RNA editing has different properties depending on affected genomic regions. For example, Alu editing is different from recoding editing. Is there a way to take such properties into account?
Finally, DRETools features should better described in the manuscript and details about input and output files should be included in the wiki pages.
Some experimental validations are required to corroborate tool results.
In my opinion DRETools are useful but major improvement is deeply needed.

Is the description of the software tool technically sound? Partly
Are sufficient details of the code, methods and analysis (if applicable) provided to allow replication of the software development and its use by others? Yes

Is sufficient information provided to allow interpretation of the expected output datasets and any results generated using the tool? Partly
Are the conclusions about the tool and its performance adequately supported by the findings presented in the article? Partly No competing interests were disclosed. Competing Interests: I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.