rnaSPAdes: a de novo transcriptome assembler and its application to RNA-Seq data.

Bushmanova E, Antipov D, Lapidus A, Prjibelski AD. · 2019

Read rnaSPAdes: a de novo transcriptome assembler and its application to RNA-Seq data. by Bushmanova E, Antipov D, Lapidus A, Prjibelski AD. (2019) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6736328.txt.

Introduction

Bushmanova and co-authors present rnaSPAdes, a tool that assembles transcriptomes from scratch. RNA sequencing yields huge datasets. Reference-based assemblers work only for species with finished genomes. Building transcripts without a reference stays hard because genes vary in activity and come in spliced forms. Similar genes add to the confusion.

RnaSPAdes builds on the SPAdes genome assembler. It treats transcriptome assembly as akin to single-cell genome assembly. The authors compared it with rival tools on many datasets and quality measures. No tool won on every count. Still, rnaSPAdes typically beat the others, with cleaner contigs and fewer duplicates. The report spells out each assembler's strong and weak points.

Catalog record

CREATORBushmanova E, Antipov D, Lapidus A, Prjibelski AD.
YEAR2019
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC6736328

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

rnaSPAdes: a de novo transcriptome assembler and its application to RNA-Seq data By Elena Bushmanova, Dmitry Antipov, Alla Lapidus, Andrey D Prjibelski Published in: GigaScience DOI: 10.1093/gigascience/giz100 ABSTRACT Abstract Background The possibility of generating large RNA-sequencing datasets has led to development of various reference-based and de novo transcriptome assemblers with their own strengths and limitations. While reference-based tools are widely used in various transcriptomic studies, their application is limited to the organisms with finished and well-annotated genomes. De novo transcriptome reconstruction from short reads remains an open challenging problem, which is complicated by the varying expression levels across different genes, alternative splicing, and paralogous genes. Results Herein we describe the novel transcriptome assembler rnaSPAdes, which has been developed on top of the SPAdes genome assembler and explores computational parallels between assembly of transcriptomes and single-cell genomes. We also present quality assessment reports for rnaSPAdes assemblies, compare it with modern transcriptome assembly tools using several evaluation approaches on various RNA-sequencing datasets, and briefly highlight strong and weak points of different assemblers. Conclusions Based on the performed comparison between different assembly methods, we infer that it is not possible to detect the absolute leader according to all quality metrics and all used datasets. However, rnaSPAdes typically outperforms other assemblers by such important property as the number of assembled genes and isoforms, and at the same time has higher accuracy statistics on average comparing to the closest competitors. BACKGROUND While reference-based methods for RNA-sequencing (RNA-Seq) analysis [1–6] are widely used in transcriptome studies, they are subjected to the following constraints: (i) they are not applicable when the genome is unknown, (ii) their performance deteriorates when the genome sequence or annotation is incomplete, and (iii) they may miss unusual transcripts such as fusion genes or genes with short unannotated exons. To address these constraints, de novo transcriptome assemblers [7–11] have emerged as a viable complement to the reference-based tools. Although de novo assemblers typically generate fewer complete transcripts than the reference-based methods for organisms with accurate reference sequences [12], they may provide additional insights on aberrant transcripts. While transcriptome assembly may seem to be a simpler problem than genome assembly, RNA-Seq assemblers have to address the complications arising from highly uneven read coverage depth caused by variations in gene expression levels. However, this is the same challenge that we have addressed while developing the SPAdes assembler [13,14], which originally aimed at single-cell sequencing. Similarly to RNA-Seq, the multiple displacement amplification (MDA) technique [15], used for genome amplification of single bacterial cells, results in a highly uneven read coverage. In view of similarities between RNA-Seq and single-cell genome assemblies, we decided to test SPAdes without any modifications on transcriptomic data. Even though SPAdes is a genome assembler and was not optimized for RNA-Seq data, in some cases it generated decent assemblies of quality comparable to the state-of-the-art transcriptome assemblers. To perform the benchmarking we have used the rnaQUAST tool [16], which was designed for quality evaluation of de novo assemblies with the support of a reference genome and its gene database. For the comparison, we selected a few representative metrics such as (i) total number of assembled transcripts (contigs), (ii) reference gene database coverage, (iii) number of 50%/95%-assembled genes/isoforms, (iv) number of misassemblies, and (v) duplication ratio. A detailed description of these metrics can be found in the Supplementary material (section S1). Table 1 compares the performance of different assembly tools on a publicly available mouse RNA-Seq dataset. All transcriptome assemblers were launched with default parameters; SPAdes was run in single-cell mode owing to the uneven coverage depth of RNA-Seq data. Table 1 shows that SPAdes generated more 50%/95%-assembled genes than any other tool and yielded comparable gene database coverage. At the same time, SPAdes produced a rather high number of misassembled transcripts, which can be explained by the fact that algorithms for genome assembly tend to assemble longer contigs and may incorrectly join sequences corresponding to different genes when working with RNA-Seq data. In addition, SPAdes generated the same number of 95%-assembled genes and isoforms, which emphasizes the lack of an isoform detection step. [Table: Benchmarking of BinPacker, Bridger, IDBA-tran, RNA-Bloom, SOAPdenovo-Trans, SPAdes, Trans-ABySS, and Trinity on mouse RNA-Seq dataset (accession No. SRX648736)] Benchmarking on other datasets also showed that SPAdes successfully deals with non-uniform coverage depth and produces a relatively high number of 50%/95%-assembled genes in most cases. However, it also confirmed the problem of a large amount of erroneous transcripts as well as relatively few fully reconstructed alternative isoforms in SPAdes assemblies. On the basis of the obtained statistics we decided to adapt current SPAdes algorithms for RNA-Seq data with the goal of improving the quality of generated assemblies and develop a new transcriptomic assembler called rnaSPAdes. In this article we describe major pipeline modifications as well as several algorithmic improvements introduced in rnaSPAdes that allow it to avoid misassemblies and obtain sequences of alternatively spliced isoforms. To perform sufficient benchmarking of rnaSPAdes and the other aforementioned transcriptome assemblers, we assembled several simulated and publicly available real RNA-Seq datasets from the organisms with various splicing complexity. For the generated assemblies we present quality assessment reports obtained with different de novo and reference-based evaluation approaches. In addition, based on these results we discuss strengths and disadvantages of various assembly tools and provide insights on their performance. DATA DESCRIPTION To compare the performance of rnaSPAdes with that of the state-of-the-art transcriptome assemblers we selected 2 simulated and 6 real publicly available RNA-Seq datasets (Table 2) with different (i) read length, (ii) library size, (iii) strand specificity, and (iv) organism splicing complexity. Simulated data were generated using the RSEM simulator [1] based on the real human and mouse datasets used in this study (the exact commands are provided in Supplementary material section S3). [Table: RNA-Seq datasets selected for comparison of different assembly tools] All downloaded public datasets were analyzed using FastQC [17]. The reports showed that no dataset contains adaptors or overrepresented sequences. The Human large dataset was quality-trimmed using Trimmomatic [18] owing to a decrease in quality towards the reads' ends. All other datasets were assembled without additional preprocessing. Although the 8 datasets used in this article may not represent all kinds of transcriptomic data, they are sufficient for comparing different assembly tools and detecting their strengths and disadvantages. ANALYSES Selected datasets were assembled with BinPacker [19], Bridger [20], IDBA-tran [10], RNA-Bloom [21], SOAPdenovo-Trans [11], Trans-ABySS [7], Trinity [8], and rnaSPAdes using default parameters, and SPAdes [13] in single-cell mode. While rnaSPAdes automatically calculates k-mer sizes based on the read length (see Methods for details), other assemblers have fixed default k-values. Indeed, varying k-value may affect the assembly in both positive and negative ways. However, because detecting the best k-mer sizes for all third-party assemblers requires additional large-scale analysis and is outside the scope of this work, it remains unclear how to properly select k for other tools. Thus, we decided to stick to default k-values, which were used in the original manuscripts or suggested in the user manuals by their developers and therefore are likely to be used by users. For a fair comparison the same minimal contig length cut-off was used for all tools (200 bp). For assemblers that have no such

Read the complete text