Source-linked AI summary
A5-miseq: an updated pipeline to assemble microbial genomes from Illumina MiSeq data
David Coil, Guillaume Jospin, Aaron E. Darling
TL;DR
Genome assembly requires multiple processing steps, and earlier A5 software was unable to process reads longer than 150nt. A5-miseq revises the pipeline for newer MiSeq data and produces improved assemblies with minimal user effort.
Problem
Genome assembly spans multiple processing steps, while the original A5 pipeline could not process MiSeq reads longer than 150nt.
Method
A5-miseq replaces components of A5 with modules for read cleaning, error correction, contig assembly, scaffolding, and misassembly correction.
Results
A5-miseq assemblies contain more full-length reference genes than A5 or GAGE-B assemblers, with 1.6 to 3.8-fold fewer missing genes than A5.
Takeaways & Limitations
A5-miseq recovers a fixed genome fraction, such as 95%, with about half as much sequence data as A5 and requires minimal user effort.
Takeaways & Limitations
GAGE-B benchmarking may yield artificially high scores because assembly pipelines could be tuned specifically to that dataset.
Abstract
from arXiv · showhide
Motivation: Open-source bacterial genome assembly remains inaccessible to many biologists due to its complexity. Few software solutions exist that are capable of automating all steps in the process of de novo genome assembly from Illumina data. Results: A5-miseq can produce high quality microbial genome assemblies from as little as 20-fold sequence data coverage on a laptop computer without any parameter tuning. A5-miseq does this by automating the process of adapter trimming, quality filtering, error correction, contig and scaffold generation, and detection of misassemblies. Unlike the original A5 pipeline, A5-miseq can use long reads from the Illumina MiSeq, use read pairing information during contig generation, and includes several improvements to read trimming. Together these changes result in substantially improved assemblies that recover a more complete set of reference genes than previous methods. Availability: A5-miseq is licensed under the GPL open source license. Source code and precompiled binaries for Mac OS X 10.6+ and Linux 2.6.15+ are available from http://sourceforge.net/projects/ngopt
5 Introduction
A5-miseq revises the original A5 pipeline to process longer MiSeq reads and produce substantially improved microbial genome assemblies. It automates the workflow from raw sequence data through scaffolded contigs and performs competitively on assembly-accuracy measures.
- Genome assembly spans adapter trimming, quality filtering, error correction, contig creation and verification, and scaffold generation.
- MiSeq reads exceeding 400nt are fourfold longer than reads previously possible on HiSeq 2000, but the original A5 could not process reads longer than 150nt.These longer reads can support assembly from less data but required major revisions to A5’s algorithms.
- A5-miseq replaces several original A5 components with new software modules and produces substantially improved assemblies.
- At 20x coverage, A5-miseq assemblies become highly contiguous, whereas A5 requires 50x coverage for comparable contiguity.The comparison uses B. cereus GAGE-B data; A5 used reads shortened from 250nt to 150nt, while A5-miseq used both shortened and full-length reads.
- Table 1 compares A5-miseq and other assemblers using assembly-accuracy measures including genome fraction, NGA50, misassemblies, mismatches, and intact genes.
6 Methods
A5-miseq automates five assembly stages, revising read cleaning and contig assembly to retain usable sequence and exploit paired-read information. Its benchmarks use GAGE-B MiSeq datasets and QUAST-based accuracy evaluation across downsampled coverage levels.
- A5-miseq performs read cleaning, contig assembly, crude scaffolding, misassembly correction, and final scaffolding.Read cleaning uses Trimmomatic and SGA; contig assembly uses IDBA-UD; misassemblies are detected from read pairs mapping at unexpected distances.
- Instead of discarding reads with any adapter readthrough, A5-miseq trims only the contaminated portion.More than 50% of reads can contain adapter readthrough in extreme Nextera XT cases with paired-end 300nt chemistry.
- IDBA-UD uses paired and unpaired reads during contig generation, reducing misassembled contigs that require later correction.The implementation was revised for Mac OS X operation and lower laptop memory usage.
- The benchmark used paired-end 250nt MiSeq reads at 100x coverage for four GAGE-B organisms, with A5 inputs shortened to 150nt.Runtime and peak memory were measured on a 2012 MacBook Air running Ubuntu 13.10.
- QUAST v2.2 measured assembly accuracy against NCBI reference assemblies and annotated genes using GAGE-compatible parameters.
- Datasets were randomly downsampled from 100x to 50x, 40x, 30x, 20x, 10x, and 5x coverage before QUAST evaluation.
7 Discussion
On GAGE-B data, A5-miseq substantially improves assembly completeness and commonly improves contiguity and error metrics relative to A5. It also reaches near-complete assemblies with less sequence data and runs on laptop hardware, though GAGE-B may favor tuned pipelines.
- A5-miseq recovered more full-length reference genes than A5 or any GAGE-B assembler, with 1.6 to 3.8-fold fewer missing genes than A5.It usually also had higher NGA50 values, fewer misassemblies, and fewer base-calling errors than A5.
- A5-miseq required about half as much sequence data as A5 to recover a fixed genome fraction such as 95%.The pipeline achieved higher contiguity with less data, permitting deeper sequencing multiplexing.
- 2.2 hours, 4GB peak memory, and 5.7GB disk usage were sufficient to assemble B. cereus GAGE-B data on a laptop.Compute requirements for other bacterial genomes were reported as similar.
- GAGE-B may produce artificially high scores if assembly pipelines were inadvertently tuned specifically to that dataset.The authors state that A5-miseq and its component programs were not tuned for GAGE-B.
8 Conclusion
A5-miseq is positioned as a particularly useful option for researchers with limited bioinformatics experience or computing resources. Its advantages include automated trimming, more complete gene recovery, NCBI-ready outputs, and base-call quality scores.
- A5-miseq should be particularly useful for researchers with limited bioinformatics experience or computing resources.
- Its stated advantages include automated adapter trimming, more full-length genes assembled, NCBI-ready outputs, and base-call quality scores.The paper also advises researchers to become acquainted with available algorithms before selecting an approach.