Abstract
Motivation
Quality control of sequencing data is the first step in many sequencing workflows. Short- and long-read sequencing technologies have many commonalities with regard to quality control. Several quality control programs exist; however, none possess a feature set that is adequate for both technologies. Quality control programs aimed at Oxford Nanopore Technologies sequencing lack vital features, such as adapter searching, overrepresented sequence analysis, and duplication analysis.
Results
Sequali was developed to provide sequencing quality control for both short- and long-read sequencing technologies. It features adapter search, overrepresented sequence analysis, and duplication analysis and supports FASTQ and uBAM inputs. It is significantly faster than comparable sequencing quality control programs for both short- and long-read sequencing technologies.
Availability and Implementation
Sequali is an open-source Python application using C extensions and is freely available under the AGPL-3.0 license at https://github.com/rhpvorderman/sequali. The source code for each release is archived at zenodo: https://zenodo.org/doi/10.5281/zenodo.10822485.
1 Introduction
Quality control of sequencing data is the first step in many sequencing workflows. Though many different sequencing technologies are available, there are many commonalities between platforms with regard to quality control. Statistics pertaining to sequence length, sequence composition, and sequence quality are relevant for all platforms. Furthermore, both long- and short-read sequencing technologies use adapter sequences and barcodes, which makes sequence content analysis necessary to determine whether these helper sequences are still present in the library. Duplication profiles are useful in both PCR and PCR-free methods as the rate of natural read duplication differs between DNA and RNA experiments (Bansal 2017).
Nanopore sequencing technology enables a wide variety of applications (Pugh 2023). Since the launch of the MinION device in 2015, various tools have been launched to assess the quality of Oxford Nanopore Technologies (ONT) sequencing data, such as NanoPlot (De Coster and Rademakers 2023), MinIONQC (Lanfear et al. 2018), and pycoQC (Leger and Leonardi 2019). These tools work from sequencing summary data and, as such, do not report on sequence composition and sequence content, nor do they provide information on the duplication profile. As a result, these tools may miss problematic libraries with adapter and barcode contaminations. Channel translocation speed deviations are an important source of insertions and deletions (Delahaye and Nicolas 2021), but NanoPlot, MinIONQC, and pycoQC do not provide any plots on translocation speed.
FastQC (Andrews 2010) and Fastp (Chen et al. 2018) are commonly used for short-read sequencing and do provide statistics for adapter and barcode contaminations. However, FastQC and Fastp do not provide proper support for ONT data. ONT adapters are not included by default, and the Q scores in their quality metrics are not based on expected error rate, as is best practice (Edgar and Flyvbjerg 2015). As a result, the average quality of reads is overestimated. Not using expected error rates to calculate the quality also limits the viability of FastQC and Fastp for short-read sequencing data. This was brought to the attention of the developers in 2023, but not fixed at the time of writing this paper.
In order to address these concerns, this paper presents a new tool for sequencing quality control: Sequali.
2 Implementation
Sequali is a collection of sequence analysis modules written in C with Python bindings. The Sequali application is written in Python and handles the overarching logic as well as the report generation. It has been made available on the Python Package Index (PyPI), Bioconda (Grüning et al. 2018), and BioContainers (Da Veiga Leprevost et al. 2017).
One module handles generic statistics, such as Phred score and base content. The adapter content module searches for small 12 bp probes for single-end data and uses a fast multiple pattern-matching algorithm (Mukherjee 2010). For paired-end, data read overlap is used to determine the adapter and insert size as this provides better detection (Turner 2014). The duplication estimation module takes a small 16 bp fingerprint from the read and uses the technique described by Xie et al. (2013) to gain an accurate estimate of the duplicated content. To check for adapter and helper sequences that are not included in the adapter search module, one in eight reads is sampled into small fragments. The first 5 million fragments are stored, and their abundance is counted in the rest of the sample. Any fragment with a higher than 0.1% occurrence in the sample is identified against the NCBI UniVEC database (NCBI 2023) and some known ONT barcodes and adapters, which are not included in UniVec yet. The identification occurs by gathering candidates using canonical k-mers and determining sequence identity using Smith–Waterman alignment. Separate modules for Illumina (per tile quality, insert size metrics) and ONT sequencing (per channel plots, translocation speed plots) are provided. A flowchart of Sequali’s internal workflow is shown in Fig. 1.
Figure 1.
Flowchart representation of Sequali’s internal workflow.
Sequali offers MultiQC (Ewels et al. 2016) integration and works from Sanger FASTQ (Cock et al. 2010) and uBAM (Li et al. 2009) formats.
To improve the speed of the application, the Python bindings for the ISA-L (Tucker et al. 2017) library are used to speed up decompression speed of the input files.
3 Methods
Only tools that can work from FASTQ data and were still actively maintained were considered for comparison. NanoPlot, FastQC, and Fastp are all popular tools that can handle the FASTQ data format. Falco is a C++ reimplementation of FastQC that vastly improves the runtime characteristics (de Sena Brandine and Smith 2019). Since Falco and FastQC effectively create the same outputs, Falco is used in the performance comparison. Since Fastp is actually also a read modification tool, all its modification options are disabled using the -A -G -Q -L flags as was done in the Falco paper (de Sena Brandine and Smith 2019). Furthermore, despite disabling all read modification options, Fastp still runs a before and after modification reporting. Fastp runtime is divided by two to correct for it running the QC twice. The Fastp runtime performance is greatly affected by turning on overrepresented sequence sampling, so it is run both with and without.
For comparing the ONT data performance, the ERR3988483 dataset from a study by Shafin et al. (2020) was used. For comparing performance on Illumina reads, the ERR11204024 dataset was used containing two paired-end FASTQ files. For tools that cannot natively handle paired-end data, the tool was run on each file, and the run times were added. Links to the benchmark files are provided on the Sequali GitHub page.
Runtime was measured by adding reported system time and user time as reported by the GNU time (GNU 2018) application. Tools were run with apptainer 1.2.5 with the—containall flag using a container provided by BioContainers (Da Veiga Leprevost et al. 2017) with only the test data directory mounted on a AMD Ryzen 5 3600 system with 2x16GB DDR4-3200 memory.
4 Results
An overview of the feature comparison is shown in Table 1. Sequali and NanoPlot both use the expected error rate to calculate the average Phred score, whilst FastQC/Falco and Fastp average the Phred scores. As a result, FastQC/Falco and Fastp consistently report much higher quality scores than NanoPlot and Sequali. For example, Falco reports over 1 000 000 reads in the ERR3988483 file to be at quality Q ≥ 20 and more than 700 000 reads to be at quality Q ≥ 30. In contrast, Sequali and NanoPlot report just 28 reads to be Q ≥ 15.
Table 1.
Feature table.
| Sequali | FastQC/Falco | Fastp | NanoPlot | |
|---|---|---|---|---|
| Length plots | Yes | Yes | Yes | Yes |
| Phred score plots | Yes | Yes, but overestimates quality | Yes, but overestimates quality | Yes |
| GC content plots | Yes | Yes | Yes | No |
| Per read averages | Yes | Yes | No | Yes |
| ONT channel plots | Yes | No | No | Yes |
| ONT translocation speed plot | Yes | No | No | No |
| Illumina per tile quality plot | Yes | Yes | No | No |
| Insert size metrics | Yes | No | Yes | No |
| Adapter content checks | Illumina, ONT | Illumina | Illumina, MGI/BGI | None |
| Overrepresented sequences check | Against UniVec database and ONT barcodes (> 6000 sequences) | Custom database (> 150 sequences) | Yes, but sequences are not identified. | No |
| Sequence duplication profile | Yes | Yes, slightly overestimates unique reads. | No, estimates total duplicates only | No, does not estimate |
FastQC, Fastp, and Sequali perform similar overrepresented sequence analyses where the first reads are sampled, and the samples are checked for occurrence in the other reads. FastQC takes only a sample from the beginning of the read, while Fastp and Sequali can sample across the entire read. Fastp stores the indices at which these samples commonly occur. This more thorough check leads to increased runtime, and the check is turned off by default. FastQC and Sequali both check overrepresented sequences against a database. All tools use subsampling to reduce the computational load.
FastQC provides slightly inaccurate estimates for duplication because taking the first 100 000 reads and counting their occurrences tends to overestimate unique reads (Xie et al. 2013). Fastp uses a bloom filter and cannot count if sequences are duplicated multiple times, therefore, omitting a duplication profile plot. Sequali uses hash-based sampling resulting in an accurate estimate whilst using little memory (Xie et al. 2013).
The results of the performance comparison are shown in Table 2. Sequali is the fastest QC program of the programs tested. Most programs have low memory usage, though Fastp does not scale well to ultra-long reads, requiring over 30 GB to finish the run.
Table 2.
Performance comparison.
| Sequali | FastQC/Falco | Fastp | NanoPlot | |
|---|---|---|---|---|
| Biocontainer tag | 11.1—py312hf67a6ed_0 | 2.2—hdcf5f25_0 (Falco) | 0.23.4—h125f33a_5 | 1.43.0—pyhdfd78af_0 |
| Runtime ERR11204024 | 0m19s | 0m35s | 0m26s/0m52sa,b | 2m41s |
| Runtime ERR3988483 | 3m22s | 17m12s | 8m39s/11m30sa,b | 23m28s |
| Memory usage ERR11204024 | 0.5G | 0.1G | 1.8G/2.2Ga | 1.0G |
| Memory usage ERR3988483 | 0.7G | 0.6G | 30.7G/30.9Ga | 1.4G |
Without/with overrepresented sequences analysis enabled.
FastP runtimes are divided by two to account for the fact that it runs the quality control modules twice.
5 Discussion
For short-read sequencing, Sequali offers an improved experience over FastQC/Falco with faster runtimes, a more extensive overrepresented sequences check that also checks the ends of reads, Phred averaging based on expected error rate, and paired-end support with insert size metrics. Fastp offers a suite of FASTQ manipulation tools, which Sequali does not offer. However, running only the QC component, Sequali is much faster whilst offering very useful statistics such as per read average quality and per read GC content plots.
Sequali offers significantly more features than ONT-oriented tools. Currently, none of the mentioned ONT-oriented tools do any analysis based on the content of the sequence. While FastQC/Falco and Fastp can technically also be used with ONT data, Sequali offers better integration by having adapter and bar code content searches for ONT-specific sequences, specific plots for channel usage, and translocation speed as well as much lower resource requirements for long-read data.
Due to its suitability for both long- and short-read data, Sequali can be used as a universal QC tool, replacing the need for having separate QC tools for ONT and Illumina data.
Acknowledgements
Simon Andrews’ FastQC application was a big inspiration for Sequali. Despite its drawbacks, it still stands as a very well-thought-out application, which is focused on determining faulty sequence libraries quickly. Shifu Chen’s Fastp inspired the use of overlap analysis. Wouter de Coster’s NanoPlot application provided inspiration for the ONT plots. Wouter de Coster also wrote an excellent blog post on how to average Phred scores. Marcel Martin was of great help by providing lots of feedback on early releases of Sequali and sharing his wealth of knowledge on application development. Colleagues Gijs van Santen and Jeroen Knijnenburg provided Oxford Nanopore sequencing samples as well as providing feedback on assessing sample quality in a clinical context, which ensured Sequali worked on a wide array of input data. Davy Cats, Ioannis Moustakas, Susan Kloet, Roberta Menafra, and Yavuz Ariyurek provided feedback on the application and its report. Davy Cats and Hailiang Mei provided feedback on the manuscript. Hailiang Mei provided support and guidance to publish the application. The reviewers comments and suggestions were very helpful for improving the manuscript and the tool documentation.
Conflict of interest
None declared.
Funding
This work has been supported by the Department of Biomedical Data Sciences and the Department of Clinical Genetics, Leiden University Medical Center.
Data availability
Sequali is freely available under the AGPL-3.0 license. The source code and additional examples can be found at https://github.com/rhpvorderman/sequali. The source code for each release is archived at zenodo: https://zenodo.org/doi/10.5281/zenodo.10822485.
Binary releases for Linux, Windows and macOS are available on PyPI (https://pypi.org/project/sequali/). Binary releases for Linux and macOS are available on Bioconda (https://bioconda.github.io/recipes/sequali/README.html). Installation can be via pip, conda or by using a container provided by the BioContainers project.
References
- Andrews S. FastQC: a quality control tool for high throughput sequence data. 2010. http://www.bioinformatics.babraham.ac.uk/projects/fastqc/ (12 August 2024, date last accessed).
- Bansal V. A computational method for estimating the PCR duplication rate in DNA and RNA-seq experiments. BMC Bioinformatics 2017;18:43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chen S, Zhou Y, Chen Y et al. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 2018;34:i884–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cock PJA, Fields CJ, Goto N et al. The Sanger FASTQ file format for sequences with quality scores, and the Solexa/Illumina FASTQ variants. Nucleic Acids Res 2010;38:1767–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Da Veiga Leprevost F, Grüning BA, Alves Aflitos S et al. BioContainers: an open-source and community-driven framework for software standardization. Bioinformatics 2017;33:2580–2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- De Coster W, Rademakers R. NanoPack2: population-scale evaluation of long-read sequencing data. Bioinformatics 2023;39:btad311. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Delahaye C, Nicolas J. Sequencing DNA with nanopores: troubles and biases. PLoS One 2021;16:e0257521. [DOI] [PMC free article] [PubMed] [Google Scholar]
- de Sena Brandine G, Smith AD. Falco: high-speed FastQC emulation for quality control of sequencing data. F1000Res 2019;8:1874. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Edgar RC, Flyvbjerg H. Error filtering, pair assembly and error correction for next-generation sequencing reads. Bioinformatics 2015;31:3476–82. [DOI] [PubMed] [Google Scholar]
- Ewels P, Magnusson M, Lundin S et al. MultiQC: summarize analysis results for multiple tools and samples in a single report. Bioinformatics 2016;32:3047–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- GNU. Time 1.9. (March 2018). https://www.gnu.org/software/time/ (26 November 2024, date last accessed).
- Grüning B, Dale R, Sjödin A et al. ; The Bioconda Team. Bioconda: sustainable and comprehensive software distribution for the life sciences. Nature Methods 2018;15:475–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lanfear R, Schalamun M, Kainer D et al. MinIONQC: fast and simple quality control for MinION sequencing data. Bioinformatics 2018;35:523–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Leger A, Leonardi T. pycoQC, interactive quality control for Oxford Nanopore Sequencing. JOSS 2019;4:1236. [Google Scholar]
- Li H, Handsaker B, Wysoker A et al. ; 1000 Genome Project Data Processing Subgroup. The Sequence Alignment/Map format and SAMtools. Bioinformatics (Oxford, England) 2009;25:2078–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Mukherjee SFA. Multiple pattern matching. In: PATTERNS 2010, The Second International Conferences on Pervasive Patterns and Applications, November 21–26, Lisbon, Portugal, pp. 78–83. IARIA, 2010.
- NCBI. UniVec 10.1. 2023. https://www.ncbi.nlm.nih.gov/tools/vecscreen/univec/ (12 August 2024, date last accessed).
- Pugh J. The current state of nanopore sequencing. In: Arakawa K (ed.), Nanopore Sequencing, Methods in Molecular Biology, Vol. 2632, New York, NY: Springer US, 2023, 3–14. [DOI] [PubMed]
- Shafin K, Pesout T, Lorig-Roach R et al. Nanopore sequencing and the Shasta toolkit enable efficient de novo assembly of eleven human genomes. Nat Biotechnol 2020;38:1044–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tucker G, Oursler R, Stern J. ISA-L Igzip: improvements to a Fast Deflate. In: 2017 Data Compression Conference (DCC), Snowbird, Utah, USA, p.465. Los Alamitos, CA: IEEE Computer Society, 2017.
- Turner FS. Assessment of insert sizes and adapter content in fastq data from NexteraXT libraries. Front Genet 2014;5:5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Xie F, Condict M, Shete S. Estimating duplication by content-based sampling. In: Proceedings of the 2013 USENIX Conference on Annual Technical Conference, USENIX ATC’13, San Jose, CA, USA, pp. 181–6, USA: USENIX Association, 2013.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Sequali is freely available under the AGPL-3.0 license. The source code and additional examples can be found at https://github.com/rhpvorderman/sequali. The source code for each release is archived at zenodo: https://zenodo.org/doi/10.5281/zenodo.10822485.
Binary releases for Linux, Windows and macOS are available on PyPI (https://pypi.org/project/sequali/). Binary releases for Linux and macOS are available on Bioconda (https://bioconda.github.io/recipes/sequali/README.html). Installation can be via pip, conda or by using a container provided by the BioContainers project.