- Open Access
Oasis 2: improved online analysis of small RNA-seq data
© The Author(s). 2018
- Received: 25 August 2017
- Accepted: 29 January 2018
- Published: 14 February 2018
Small RNA molecules play important roles in many biological processes and their dysregulation or dysfunction can cause disease. The current method of choice for genome-wide sRNA expression profiling is deep sequencing.
Here we present Oasis 2, which is a new main release of the Oasis web application for the detection, differential expression, and classification of small RNAs in deep sequencing data. Compared to its predecessor Oasis, Oasis 2 features a novel and speed-optimized sRNA detection module that supports the identification of small RNAs in any organism with higher accuracy. Next to the improved detection of small RNAs in a target organism, the software now also recognizes potential cross-species miRNAs and viral and bacterial sRNAs in infected samples. In addition, novel miRNAs can now be queried and visualized interactively, providing essential information for over 700 high-quality miRNA predictions across 14 organisms. Robust biomarker signatures can now be obtained using the novel enhanced classification module.
Oasis 2 enables biologists and medical researchers to rapidly analyze and query small RNA deep sequencing data with improved precision, recall, and speed, in an interactive and user-friendly environment.
Availability and Implementation
Small RNAs (sRNAs) are a class of short, non-coding RNAs with important biological functions in nearly all aspects of organismal development in health and disease. Especially in diagnostic and therapeutic research sRNAs, such as miRNAs and piRNAs, received recent attention . The current method of choice for the quantification of the genome-wide sRNA expression landscape is deep sequencing (sRNA-seq).
To date several local as well as server-based sRNA-seq analysis workflows are available that differ in their analysis portfolio, performance, and user-friendliness. Analysis workflows that need to be installed by the end-user comprise, for example, sRNA workbench  for the quantification and identification of differentially expressed sRNAs and CAP-miRSeq  for the quantification of known and novel miRNAs including variant calling and subsequent differential expression analysis. While workflows that are installed on a local machine offer greater data security and may provide greater flexibility, they require installation, availability of servers, software and hardware maintenance as well as regular updates.
sRNA-seq web application comparison
miRNA modifications and edits
Novel miRNA database
Infection and cross-species analysis
Multivariate differential expression
Novel miRNA target prediction
Batch job submission (API)
At the heart of Oasis 2 lies the new sRNA detection workflow that is faster and identifies more sRNAs with higher precision. In addition, Oasis 2 now supports sRNA-seq analyses for any organism, detects potential cross-species miRNAs, and reports viral and bacterial infections in samples with high precision and recall. Oasis 2 predicts and stores novel miRNAs in Oasis-DB and allows users to search and extract information for over 700 predicted high-quality miRNAs across 14 organisms. Oasis 2 classification module is improved with the use of balanced sampling and feature pruning methods that enables robust biomarker detection. Like its predecessor Oasis, Oasis 2’s differential expression module supports multiple group comparisons (e.g. control vs. treatment 1 vs. treatment 2) and differential expression using co-variates such as age, gender, and medication. The differential expression and classification modules report various quality metrics including known and predicted targets of miRNAs in a downloadable, interactive web report. This web report allows for the subsequent functional enrichment analysis of miRNAs using GeneMania (interactome and GO analysis) , g:Profiler (GO, pathway-Kegg, Reactome) , STRING (protein-protein interaction network) , STITCH (chemical-protein interaction network) , and DAVID (enrichment analysis based on many biological databases) . Oasis 2 is also at the heart of the sRNA Expression Atlas (SEA, https://sea.dzne.de), a web application for the interactive querying, visualization, and analysis for over 2000 published sRNA samples. Lastly Oasis 2 features many new analysis and visualization options such as support for adapter trimmed data, options to trim additional barcodes, and interactive plots for sRNA detection and classification output. It has no restrictions on the size or number of samples and has no limits on the analyses per user.
The following paragraphs will describe the technical details of Oasis 2’s novel sRNA detection, database, and classification modules. Additional information can be found in the supplementary material.
In addition to the new alignment strategy, the sRNA detection module also supports data with already trimmed adapters. It also has an option for barcode removal, which is required for the analysis of libraries generated with e.g. the NEXTflex kit. In the case of barcode removal, Oasis 2 first discards the 3′ adapter sequence (in case the adapter is not already trimmed), and then removes an additional N (user defined, default is 0) bases from the adapter-clipped reads.
Detection and storage of novel miRNAs
Another major improvement of Oasis 2 is the ability to query and visualize detailed information for over 700 high-quality predicted miRNAs across 14 organisms (Fig. 1, Additional file 1: Figure S1). Oasis-DB comprises information on all MiRDeep2  predicted miRNAs that pass stringent selection criteria during the sRNA detection step of Oasis 2 (2.1 & Additional file 1), including the miRNA ID, organism, chromosomal location, precursor and mature sequences, structure, read counts, prediction scores, and detailed information on the software and its versions used to predict the miRNA. To assure that Oasis-DB contains only high-quality miRNA entries, novel predicted miRNAs have to pass the three criteria. The log-odds score assigned to the hairpin by miRDeep2 (miRDeep2-score) should be greater than 10, the predicted miRNA hairpin should not have sequence similarity to reference tRNAs or rRNAs, and the estimated randfold p-value of the excised potential miRNA hairpin should be equal to or lower than 0.05.
Novel predicted miRNAs are added to Oasis-DB using the standard nomenclature (Additional file 1 section 1.4 ‘Oasis-DB miRNA insertion and naming’).
In addition to novel miRNAs, Oasis-DB also stores information on all other sRNAs and sRNA families (Additonal file 1). To provide access to Oasis-DB we created a novel web frontend, the Oasis 2 ‘Search’ module, which allows users to query miRNAs by mature/precursor ID or sequence, and the organism they come from. Information on high-confidence novel miRNAs is also shared with SEA, a web application that provides expression information of known and novel miRNAs for over 2000 samples (https://sea.dzne.de).
Classification and differential expression
To allow for enhanced sRNA-based biomarker detection several profound changes to the Oasis 2 classification module were made, resulting in more robust biomarker detection with increased accuracy (Additional file 1: Figure S2 , Additional file 1 section ‘Oasis 2 classification module’). To increase the performance of the Random Forest-based (RF) classification module we first implemented balanced sampling (Additional file 1), making sure RF predictions would not be biased in the case of uneven class distribution. Since RFs can perform poorly on data that contains few informative and many non-informative features, the classification module was augmented with a feature pruning routine (Additional file 1), reporting prediction performance for the full and best RF models. In addition to providing information on model accuracy using the out-of-bag (OOB) error, Oasis 2 now also provides model performance information based on cross-validation. All classification results can be explored in interactive web reports, allowing for a detailed quality and performance analysis of the predicted biomarkers.
Moreover, we have improved the quality of output plots in the DE module and updated the DESeq2 version for the analysis of differential sRNA expression. Further details about DE module can be found in Additional file 1 section 1.5 ‘Oasis 2 differential expression module’ and Additional file 1: Table S3.
Technologies and compatibility
Oasis 2 browser compatibility
55.0.3, 56.0 (64-bit), 57.0 (64-bit)
Runtime comparison of different sRNA-seq web applications
Oasis 2 (total) 1
Detection and differential expression of sRNAs
To estimate if the novel sRNA detection workflow of Oasis 2 identifies and quantifies sRNAs correctly we analyzed four published datasets containing validated sRNA changes using Oasis 2 with default settings. Of note, none of the above-mentioned publications looked into the DE of other small RNA classes (snRNA, snoRNA and rRNA and piRNAs), so the analyses were restricted to miRNAs.
Alzheimer disease data
We started by analyzing an Alzheimer disease (AD) sRNA dataset that consists of 48 Alzheimer and 22 control samples  using Oasis 2 and default settings. The original publication uses a Wilcoxon-Mann-Whitney test detecting 125 known DE miRNAs. Oasis 2 detected 103 DE miRNAs using an adjusted p-value < 0.1, of which 62(60%) overlapped with the original analysis. The overlap of 60% seems reasonable, given the different statistical approaches and miRBase versions used for the detection and DE analysis of the miRNAs. In the original publication 8/10 known miRNAs were validated to be differentially expressed in the same direction, whereas two miRNAs (hsa-miR-1285-5p and hsa-miR-26a-5p) were not validated in the same direction (instead of upregulation they showed downregulation in qPCR). Interestingly these two miRNAs were not detected to be differentially expressed by Oasis 2. On the other hand Oasis 2 was able to detect 3/3 upregulated miRNAs (hsa-let-7d-3p, hsa-miR-5010-3p and hsa-miR-151a-3p), 3/5 downregulated miRNAs (hsa-miR-532-5p, hsa-miR-26b-5p and hsa-let-7f-5p), and it did not detect two downregulated miRNAs (hsa-miR-103a-3p, hsa-miR-107). In summary, Oasis 2 was able to detect 6/8 (75%) validated differentially expressed known miRNAs and not detecting 2/2 false positives from the original study. Unfortunately, two novel miRNAs validated in the original study are not added to miRBase yet, therefore we were not able to compare to them.
Oasis 2’s performance was next assessed using a set of 10 Psoriasis and 10 control samples . The original publication uses a hypergeometric test to assess differential expression (Pearson’s chi-square test) that is followed by a Bonferroni multiple-testing correction.
Overlap of differentially expressed sRNAs using three datasets
DESeq2 (Dejian et al., 2015)
Renal cancer data
In this work 11 renal cancer and 11 remission samples  were analyzed. This is longitudinal data from 11 patients and as such paired but we were unable to extract the pairing information from the GEO database annotations. Therefore the data was analyzed with Oasis 2 in un-paired mode and compared to the published, paired analysis with edgeR . Despite of these technical issues the two analyses showed high overlap. Oasis 2 found 150 DE miRNAs (adjusted p-value < 0.1) whereas the original publication lists only 70 DE miRNAs. Of these 70 DE miRNAs 53 (76%) could also be found in the significant Oasis 2 miRNAs (Table 4). Of note, with the exception of miR-122 all the validated miRNAs from the original work were detected using Oasis 2 (miR-21-5p, miR-210-3p, miR-199, miR-532-3p).
Schizophrenia and schizoaffective disorder data
In this experiment induced pluripotent stem cells were used to study neuropsychiatric disorders associated with 22q11.2 microdeletions . Controls and patients with 22q11.2 microdeletions diagnosed with a psychotic disorder were compared (9 controls and 7 patients). Oasis 2 found 34 DE miRNAs (adjusted p-value < 0.1) whereas the original publication identified 45 DE miRNAs. Of these 45 DE miRNAs 14 (41%) were also detected as differentially expressed by Oasis 2 (Table 4). In the original publication four miRNAs were validated by qPCR, two significantly up-regulated (miR-23a-5p and miR-146b-3p), one significantly down-regulated (miR-185-5p), and a miRNA that showed no difference in expression (miR-767-5p). Oasis 2 was able to confirm 2/3 (67%) validated differentially expressed miRNAs (miR-23a-5p and miR-185-5p) and did not confirm 1/1 (100%) false positive miRNAs miR-767-5p.
Overall, Oasis 2 detected 19/27 (70%) independently validated DE miRNAs in the published datasets despite of the different statistical approaches and miRBase versions used (Table 4). Detailed analysis results are accessible in Oasis 2’s ‘Demo Data’ webpage. Our results provide strong evidence that Oasis 2 provides biologically meaningful results to the end user.
Pathogen detection and sample classification
To assess the performance of the pathogen detection we analyzed 5 datasets with known viral or bacterial infections (Additional file 1: Table S6). We calculated the precision, recall, and F-score for the detection of the particular pathogen strain in the dataset while considering only the top ranking, first two, three, and up to the first ten reported species (Fig. 2). Species were ordered based on the number of read counts. In general, the viral or bacterial species and strains were detected with high precision and recall, reaching F-scores of ~ 0.8 when the top five viral and bacterial species were considered. In consequence, Oasis 2 currently reports the top five bacterial, archaeal, and viral species found, allowing for the detection of potential infective agents or the discovery of experimental sample contaminations.
To benchmark the improved classification routine, we compared the performance of the old Oasis classification module (unbalanced sampling with all variables) to the new Oasis 2 classification module using balanced sampling and feature optimization using three demo datasets (see Detection and Differential Expression of sRNAs and Additional file 1: Figure S2). From a theoretical perspective, balanced sampling should increase prediction accuracy only in the case of class imbalances. In consequence, the novel classification module enhances the AUC for the imbalanced AD (22 controls, 48 patients) demo dataset by 2% (old AUC 0.95, new AUC 0.97), while it marginally changes classification performance for the balanced Psoriasis (10 control and 10 Psoriasis samples) (old AUC 0.90, new AUC 0.91) and Renal carcinoma (11 control and 11 cancer samples) (new and old AUC 1.00) data. Feature pruning should be crucial when a dataset contains a lot of uninformative features and very few informative features. To this end we have taken an unpublished dataset (6 controls, 6 treatments) that contains at least one feature that perfectly separates the two classes but otherwise contains mostly uninformative features. Whereas the old classification module reaches an AUC of 0 on this dataset, the new module reaches an AUC of 0.833.
Moreover, we also compared the accuracy of the new Oasis 2 classification module on the AD dataset to the published accuracy in the original manuscript . Unfortunately, we were unable to obtain the primary output of the SVM and could not follow the post-processing steps of the machine learning results as performed in the original publication (e.g. removal of miRNAs that also occur in other diseases). In brief, the original publication provides a biomarker signature of 12 miRNAs (10 annotated and two novel) that reaches an average accuracy of 80%. The Oasis 2 classification reaches an accuracy of ~ 87% (AUC of 0.97) using 320 features (no preprocessing for other diseases) and has an out-of-bag error of ~ 10%. Two miRNAs in the original paper list (has-miR-151a-3p, hsa-let-7f-5p) were also found in the top 10 features (miRNAs) obtained with Oasis 2 classification.
The classification analysis of the three demo datasets (see 3.1) yielded stable and robust biomarker predictions that further corroborated the quality of the enhanced classification module.
We next estimated the runtime of Oasis 2 using the above-mentioned AD, Psoriasis, and Renal cancer datasets and compared the results to runtime estimates for omiRas, mirTools 2.0, MAGI, Chimira and sRNAtoolbox, five recently developed web applications for the analysis of sRNA-seq data (Table 3, Additional file 1: Table S7). Performances of the sRNA Detection, DE Analysis, and Classification modules were measured on the Oasis 2 server. For benchmarking the Oasis 2 runtime we compared it to the runtime estimates of the above-mentioned web applications by submitting the AD, Psoriasis, and Renal Cancer datasets to the respective services (Table 3). Of note, runtime estimates for MAGI were taken from the MAGI webpage, which we assume constitutes a ‘best case scenario’ in favor of MAGI (low server analysis load). In addition, we could not compare to mirTools 2.0 as the maximum upload file size is limited to 30 Mb. Furthermore, the sRNAtoolbox web application was also not accessible during the period of testing and writing this manuscript.
Overall, Oasis 2 is significantly faster than MAGI, Chimira, and omiRas. For the smallest dataset (Renal Cancer) Oasis 2 was ~ 1.5 times faster than Chimira, ~ 15 times faster than MAGI, and ~ 18 times faster than omiRas. While the runtime differences between Oasis 2 and Chimira were rather small when only few samples were analyzed, Oasis 2 was ~ 2 times faster than Chimira, ~ 30 times faster than MAGI for the 48 Gb Psoriasis dataset. Unfortunately, we were unable to estimate the runtime of omiRas for the Renal Cancer dataset since it did not finish file upload. Oasis 2 analyzed the largest dataset (AD, 287 Gb) in 8 h31m50s while none of the other tools mentioned above supported the analysis of the AD samples. In summary, Oasis 2 is the fastest of the state-of-the-art web applications we could compare to and has no restrictions on the sample number or size.
Oasis 2 is fast, reliable, and offers several unique features that make it a valuable addition to the ever-growing number of sRNA-seq analysis applications. Especially the analysis support for all organisms, the detection and storage of novel miRNAs, the differential expression and classification modules, and the interactive results visualization supporting GO and pathway enrichment analyses enable biologists and medical researchers to quickly analyze, visualize, and scrutinize their data. Oasis 2 also offers rich per experiment and per sample quality control, which might be one of the most important steps in the initial data analysis. The utility of a good quality control is exemplified in the analysis of the Psoriasis dataset, which seems to contain a mis-labelled (SRR330866_PP) and an outlier (SRR330860_PP) sample (Fig. 3). The removal of the outlier and mis-labelled samples in the Psoriasis dataset increased the number of significantly DE miRNAs from 195 to 256 cases and increased the classification accuracy for the same dataset from AUC of 0.9 to 1. We would like to emphasize that this data was already analyzed in two publications and to our knowledge this is the first time that these ‘problematic’ samples were detected, providing strong evidence for the utility of Oasis 2’ QC plots. Additionally the modular structure of Oasis 2 (sRNA detection, DE and classification) makes this task even easier, as the user can run only DE (without outliers) rather than going through the sRNA detection step again. In addition Oasis 2 provides PDF and video tutorials that explain its usage and details on how to interpret its results. Future developments will include the detection of small RNA editing, modification, and mutation events as well as more detailed reports on bacterial and viral infections and contaminations.
We would like to thank Ashish Rajput, Ting Sun, Vikas Bansal, Michel Edwar Mickael, the DZNE IT, and all of the Oasis users for helpful suggestions.
This work was supported by the DFG (BO4224/4–1), the Network of Centres of Excellence in Neurodegeneration (CoEN) initiative, the Volkswagen Stiftung (Az88705), iMed – the Helmholtz Initiative on Personalized Medicine, and the BMBF grant Integrative Data Semantics in Neurodegeneration (031L0029B, IDSN).
Availability of data and materials
SB initiated the study and designed the web application as well as analyses together with RR. RR and AG designed the Oasis-DB to store novel predicted miRNA. MF enhanced the classification module. JB and VC worked on the backend implementations of different modules. AS analyzed sRNA-seq data on different web servers to benchmark Oasis 2. DSM and OS worked the interactive user interface and tutorials. All authors read and approved the final manuscript.
Ethics approval and consent to participate
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Beckers, et al. Comprehensive processing of high-throughput small RNA sequencing data including quality checking, normalization, and differential expression analysis using the UEA sRNA Workbench. RNA. 2017;823–35.Google Scholar
- Capece V, et al. Oasis: online analysis of small RNA deep sequencing data. Bioinformatics. 2015;31:2205–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Dejian, et al. MicroRNA Profiling of Neurons Generated Using Induced Pluripotent Stem Cells Derived from Patients with Schizophrenia and Schizoaffective Disorder, and 22q11.2 Del. plosone. 2015.Google Scholar
- Franceschini, et al. STRING v9.1: protein-protein interaction networks, with increased coverage and integration. Nucleic Acids Res. 2013;D808-15.Google Scholar
- Friedländer MR, et al. MiRDeep2 accurately identifies known and hundreds of novel microRNA genes in seven animal clades. Nucleic Acids Res. 2012;40:37–52.View ArticlePubMedGoogle Scholar
- Huang, et al. The DAVID Gene Functional Classification Tool: a novel biological module-centric algorithm to functionally analyze large gene lists. Genome Biol. 2007;8:R183.Google Scholar
- Joyce CE, et al. Deep sequencing of small RNAs from human skin reveals major alterations in the psoriasis miRNAome. Hum Mol Genet. 2011;20:4025–40.View ArticlePubMedPubMed CentralGoogle Scholar
- Kim J, et al. MAGI: a node.Js web service for fast microRNA-Seq analysis in a GPU infrastructure. Bioinformatics. 2014;30:2826–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Kuhn, et al. STITCH 4: Integration of protein-chemical interactions with user data. Nucleic Acids Res. 2014;D401-7.Google Scholar
- Leidinger P, et al. A blood based 12-miRNA signature of Alzheimer disease patients. Genome Biol. 2013;14:R78.View ArticlePubMedPubMed CentralGoogle Scholar
- Müller, et al. omiRas: a Web server for differential expression analysis of miRNAs derived from small RNASeq data. Bioinformatics. 2013;2651–2.Google Scholar
- Osanto S, et al. Genome-wide microRNA expression analysis of clear cell renal cell carcinoma by next generation deep sequencing. PLoS One. 2012;7Google Scholar
- Reimand, et al. G:Profiler - A web server for functional interpretation of gene lists (2011 update). Nucleic Acids Res. 2011;W307-15.Google Scholar
- Robinson MD, et al. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010;26:139–40.View ArticlePubMedGoogle Scholar
- Rueda, et al. sRNAtoolbox: an integrated collection of small RNA research tools. Nucleic Acids Res. 2015;W467–W473.Google Scholar
- Sun, et al. CAP-miRSeq: a comprehensive analysis pipeline for microRNA sequencing data. BMC Genomics. 2014;15:423.Google Scholar
- Vitsios DM, Enright AJ. Chimira: analysis of small RNA sequencing data and microRNA modifications. Bioinformatics. 2015;31:3365–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Witwer KW. Circulating MicroRNA biomarker studies: pitfalls and potential solutions. Clin Chem. 2014;000Google Scholar
- Wood DE, Salzberg SL. Kraken: ultrafast metagenomic sequence classification using exact alignments. Genome Biol. 2014;15:R46.View ArticlePubMedPubMed CentralGoogle Scholar
- Wu, et al. mirTools 2.0 for non-coding RNA discovery, profiling, and functional annotation based on highthroughput sequencing. RNA Biol. 2013;1087–92.Google Scholar
- Zuberi, et al. GeneMANIA prediction server 2013 update. Nucleic Acids Res. 2013;W115-22.Google Scholar