Source-linked AI summary
SOAPdenovo-Trans: De novo transcriptome assembly with short RNA-Seq reads
Yinlong Xie, Gengxiong Wu, Jingbo Tang, Ruibang Luo, Jordan Patterson, Shanlin Liu, Weihua Huang, Guangzhu He, Shengchang Gu, Shengkang Li, Xin Zhou, Tak-Wah Lam, Yingrui Li, Xun Xu, Gane Ka-Shu Wong, Jun Wang
TL;DR
Short RNA-Seq reads make complete de novo transcript recovery difficult, especially when expression levels vary and alternative splicing creates complex graph structures. SOAPdenovo-Trans adapts de Bruijn-graph assembly with transcriptome-specific error handling and graph traversal, and benchmarks on rice and mouse report higher contiguity, lower redundancy, and faster execution than prior de novo assemblers. The study also notes that reference-based Cufflinks achieved better results in its test, leaving the extent of reference-free recovery uncertain.
Problem
Short RNA-Seq reads and transcriptome-specific variation make de novo recovery of complete or full-length transcripts algorithmically challenging.
Method
SOAPdenovo-Trans combines de Bruijn-graph assembly with weak global and percentage-based local error removal, transcript-specific scaffolding, and graph traversal for alternative splice forms.
Results
SOAPdenovo-Trans produced higher contiguity, lower redundancy, and faster execution than the evaluated de novo assemblers on rice and mouse transcriptome benchmarks.
Takeaways & Limitations
SOAPdenovo-Trans provides a reference-free assembler that handles alternative splicing and variable expression while recovering fewer, less redundant transcripts.
Takeaways & Limitations
Rice transcripts came from indica while the benchmark genome and annotations were japonica, so local subspecies variation prevented some transcripts from aligning.
Abstract
from arXiv · showhide
Motivation: Transcriptome sequencing has long been the favored method for quickly and inexpensively obtaining the sequences for a large number of genes from an organism with no reference genome. With the rapidly increasing throughputs and decreasing costs of next generation sequencing, RNA-Seq has gained in popularity; but given the typically short reads (e.g. 2 x 90 bp paired ends) of this technol- ogy, de novo assembly to recover complete or full-length transcript sequences remains an algorithmic challenge. Results: We present SOAPdenovo-Trans, a de novo transcriptome assembler designed specifically for RNA-Seq. Its performance was evaluated on transcriptome datasets from rice and mouse. Using the known transcripts from these well-annotated genomes (sequenced a decade ago) as our benchmark, we assessed how SOAPdenovo- Trans and two other popular software handle the practical issues of alternative splicing and variable expression levels. Our conclusion is that SOAPdenovo-Trans provides higher contiguity, lower redundancy, and faster execution. Availability and Implementation: Source code and user manual are at http://sourceforge.net/projects/soapdenovotrans/ Contact: xieyl@genomics.cn or bgi-soap@googlegroups.com
1 INTRODUCTION
RNA-Seq enables transcript-sequence discovery without a reference genome, but short reads and transcriptome-specific complexity make de novo recovery challenging. SOAPdenovo-Trans addresses these challenges and reports higher contiguity, lower redundancy, and faster execution than other evaluated assemblers.
- RNA-Seq can provide transcript sequences and gene-expression information, including allelic-expression discrimination and gene-fusion detection.
- Reference-based assemblers recover transcripts by aligning reads to a reference genome, whereas de novo assemblers use de Bruijn graphs.
- Transcriptome assembly differs from genome assembly because it must recover tens of thousands of short RNA sequences rather than tens of chromosomes.
- Expression levels can vary by many orders of magnitude, causing highly expressed genes to accumulate discrepant bases and complicating fixed error thresholds.
- SOAPdenovo-Trans combines prior algorithmic innovations with practical changes and reports higher contiguity, lower redundancy, and faster execution.The evaluation used rice and mouse transcriptome data with known transcripts from well-annotated genomes as benchmarks.
2 METHODS
SOAPdenovo-Trans adapts de Bruijn-graph assembly and scaffolding for transcriptomes by handling variable expression, alternative splicing, and short-read linkage information. Its workflow assembles contigs, builds transcript-specific scaffold graphs, traverses reliable paths, and fills gaps using paired-end reads.
- Core design: SOAPdenovo-Trans is a de Bruijn graph-based de novo transcriptome assembler derived from SOAPdenovo2 but adapted for RNA-Seq.SOAPdenovo2's genome-oriented error model and lack of alternative-splicing support were addressed by adopting and improving concepts from Trinity and Oases.
- Contig assembly: The algorithm constructs de Bruijn graphs from reads, removes sequencing errors, and generates contigs before transcript assembly.The overview identifies contig assembly as the first major stage.
- Contig assembly: Global error removal uses a weak default depth cutoff of ≤2, while local filtering uses a percentage threshold to accommodate expression-level variation.The local threshold can be set to ≤5% of the total or maximal depth of an adjacent graph element.
- Transcript assembly: Single-end and paired-end reads are mapped to contigs to build weighted linkages, with paired-end insert sizes estimating distances between linkages.Using single-end information preserves evidence that may otherwise be lost because transcripts are shorter than chromosomes.
- Transcript assembly: Ambiguous or exceptionally high-depth contigs are not simply masked because alternative splicing and variable expression are legitimate transcriptome features.The genome-oriented masking strategy is described as inappropriate for transcriptome assembly.
- Transcript assembly: A stringent linearization method simplifies transcript scaffold graphs while preserving conflicts arising from alternative splice forms.The method differs from genome assembly because multiple splice forms can legitimately share a starting contig.
- Transcript assembly: Contigs are clustered into linkage-based subgraphs, which are traversed through linear, fork, and bubble paths to generate possible transcripts.For complex paths, only the top-scoring transcripts are retained.
- Gap filling: Paired-end reads are locally assembled into consensus sequences to fill gaps introduced by masking contigs ≤100 bp before scaffold construction.When multiple consensus sequences are produced, the sequence whose length best matches the gap size is selected.
3 RESULTS
The evaluation compares transcriptome assemblers on rice and mouse, emphasizing recovery, redundancy, completeness, and putative alternative splicing. SOAPdenovo-Trans recovers similar annotated loci with fewer transcripts, approaches the ideal completeness profile more closely, and avoids some apparent isoform redundancy caused by assembly artefacts.
- Assembly output: SOAPdenovo-Trans produced fewer transcripts while recovering a similar number of annotated genome loci across algorithms.After removing assemblies below 300 bp, its transcript count was more than twofold lower in the extreme case, whereas recovered loci differed by only a few percent.
- Evaluation framework: The analysis used curated rice and mouse annotations, with series-A including and series-B excluding putative alternative splice forms.Assemblies were progressively filtered by length, alignment quality, annotated-gene overlap, and treatment of multiple assemblies at one locus.
- Dataset comparison: Rice assemblies recovered 42% to 70% more genes than mouse from similar input amounts, leading to lower read depths per rice gene.The rice–mouse difference is attributed to recovering more genes from the same data, not to transposable elements; close to 10% of rice assemblies also failed to align because of subspecies differences.
- Transcript completeness: SOAPdenovo-Trans approached the ideal full-length recovery curve more closely than Trinity or Oases.The ideal profile rises sharply near 100% of annotated length; the comparison’s shoulder suggested SOAPdenovo-Trans with 1.83 Gbp of mouse data could outperform Trinity with 5.49 Gbp.
- Isoform recovery: Using annotated isoforms as the benchmark, Trinity and Oases did not recover more isoforms than SOAPdenovo-Trans.The number of recovered isoforms depends on an arbitrary threshold such as 100% or 95% of expected length, and the benchmark excludes undiscovered isoforms.
- Alternative splicing: Many putative alternative splice forms were instead non-overlapping fragments or redundant transcripts produced by minor substitutional errors.Submaximal-to-maximal overlap ratios near one indicate no new exons, while genuine splice forms would commonly change transcript length by roughly 10 to 20%.
4 DISCUSSION
SOAPdenovo-Trans combines insights from prior assemblers and genome assembly to improve de novo transcriptome assembly. The discussion highlights both its reported superiority over previous efforts and uncertainty about how much reference-based gains can be reproduced without a genome.
- SOAPdenovo-Trans combines innovations from Trinity, Oases, SOAPdenovo2, and additional practical insights into one transcriptome assembly algorithm.The approach includes Trinity’s error removal, Oases’s graph traversal, and stricter transitive reduction for scaffolding graphs.
- The authors report SOAPdenovo-Trans as demonstrably superior to previous de novo transcriptome assembly efforts.
- Reference-based Cufflinks achieved even better results than SOAPdenovo-Trans in the authors’ tests.
- How much of the reference-based improvement can be replicated without a reference genome remains unclear.
- Multiple k-mer strategies may improve transcriptome assembly, but merging assemblies can introduce redundancy and longer reads could radically change the problem.