Troubleshooting and FAQs

Which sequencing read QC tool should I use?

nf-core/taxprofiler offers two options for quality control reporting of sequencing reads, FastQC and Falco. These tools provided various overview statistics on the sequencing quality of the reads inside your FASTQ files, allowing you to identify potential sequencing artefacts.

Falco is a recent, more performant C++, implementation of the Java-based FastQC, while producing the same output files. Falco is starting to add additional features, but is not as widely used as FastQC.

We offer FastQC as the gold-standard and most used option within the genomics community. Falco is included as a faster alternative for users who wish to try it out or need to save time and resources.

Which short read-adapter clipping/read merging tool should I use?

nf-core/taxprofiler offers two options for short-read adapter clipping and paired-end read merging, fastp and AdapterRemoval (v2). These tools provide various types of sequencing read cleanup, including removal of sequencing library adapters, low quality bases, and so on.

The two tools are both well-built tools, and offered based on popularity by different communities. For example, AdapterRemoval (v2) is often used in ancient DNA studies due to additional functionality for additional barcode data, while fastp is commonly used in modern metagenomics studies with additional general sequencing data cleanup functionality.

Each have their own advantages and disadvantages, and the choice of which to use is up to the user. You should evaluate both tools, through reading their respective documentation, literature review (such as benchmarking studies), and/or by testing on your own data.

Which long-read trimming tool should I use?

nf-core/taxprofiler offers two options for Oxford Nanopore Technologies (ONT) data long-read adapter clipping and read trimming, Porechop, and Porechop_ABI. These tools provide various types of sequencing read cleanup, including removal of sequencing library adapters, low quality bases etc.

The two tools are very similar, with the main difference being that Porechop_ABI is a more recent fork of the original Porechop, with the same functionality but without the need to pre-define adapter sequences.

Selection of the tool should be based on reading the documentation and comparing the features to the structure of your own data.

Which ONT long-read length filtering tool should I use?

nf-core/taxprofiler offers two options for Oxford Nanopore Technologies (ONT) long-read length filtering, nanoq and Filtlong. These tools provide additional quality control by removing potentially lower-quality shorter reads.

The two tools are designed to perform the same task of filtering ONT data long reads.

nanoq was developed as a more actively developed and a more performant alternative to Filtlong. It is written in Rust, with performance reported to be much faster and more efficient than the older Filtlong tool. However, Filtlong has started to be slightly more active again.

Each have their own advantages and disadvantages, and the choice of which to use is up to the user. You should evaluate both tools, through reading their respective documentation, literature review (such as benchmarking studies), and/or by testing on your own data.

Which short-read complexity filtering tool should I use?

nf-core/taxprofiler offers three options for short-read complexity filtering, PRINSEQ++, BBDuk, and fastp. Short-read complexity filtering allows you to remove non-specific low-complexity reads (such as mononucleotide repeats) that may be uninformative and slow down classification.

PRINSEQ++ is a recent and more efficient C++ version of the original PRINSEQ tool that was written in Perl. BBDuk is a well-established and widely used Java based toolkit with a large number of features and a strong user base. fastp is a general purpose sequencing QC tool that provides complexity filtering as an integrated step of read-merging and adapter trimming.

The three tools all perform well, and are offered as each provide different algorithms for complexity filtering.

• PRINSEQ++: filter based on an estimate of sequence randomness (entropy) using a DUST score.
• BBDuk: filter based on k-mer frequency
• fastp: also includes a complexity filtering step integrated into its own adapter-trimming steps

Each have their own advantages and disadvantages, with different tuning parameters and the choice of which to use is up to the user. You should evaluate all three tools, through reading their respective documentation, literature review (such as benchmarking studies), and/or by testing on your own data.

Which short-read host-removal tool should I use?

nf-core/taxprofiler offers three different tools for host-removal of short-reads.

• Bowtie2
• Hostile
• Deacon

Host-read removal can help speed up classification and reduce false-positive results. This is through removing reads that may be classified against contaminated genomes in a classification database, and/or in the case of microbial only studies, reducing the number of reads needing to be classified through the removal of uninformative reads.

Hostile and Bowtie2 provide alignment-based host-removal, while Deacon provides a k-mer based approach. If you wish to use an alignment approach for the removal of host DNA, and are working with Human-associated or common-model organism samples, we recommend using Hostile over Bowtie2. Hostile used Bowtie2 internally but as it provides optimised alignment presets as well as superior reference genomes (such as the T2T CHM13v2.0 human genome).

Deacon is a newer k-mer based approach that is reportedly much faster and efficient than alignment-based methods, with the efficiency improvements allowing the use of more inclusive pan-genomes for host-removal.

We recommend evaluating between Hostile and Deacon through reading their respective documentation, literature review (such as benchmarking studies), and/or by testing on your own data.

Which classifier(s) should I use?

Choosing a classifier is not easy! There are many pros and cons to each, and the best choice will depend on your research question, the information you need, and the trade-offs needed regarding resources (amongst others). nf-core/taxprofiler is designed to offer multiple classifiers; facilitating easier execution of the one(s) that best suit your needs.

In most cases, you should read the publication of each classifier to understand their trade-offs, reading benchmarking studies, and reviewing what other researchers are using in your specific domain.

For example, for more detailed and regularly-updated quantitative comparisons, we highly recommend seeing the latest results of automated benchmarking platforms such as:

• CAMI challenge
• LEMMI

Below we provide a brief summary of some of the characteristics of the classifiers available in nf-core/taxprofiler. With this information we hope to help you make an informed decision on which classifier(s) to use for your particular use case analysis.

Decision tree

The decision tree below can help you initially narrow down the classifiers that match your data. Kraken2 can use both DNA-to-DNA and DNA-to-AA databases, and Bracken follows Kraken2.

Full details

The table below contains all criteria that we find relevant to compare.

Furthermore, the table below provides a summary of the original primary ‘unique characteristics’ of the particular classifier (as described in the original publication of the classifier). We order by original publication year as often new-classifier ‘improvements’ are referring to earlier comparable classifiers; but may not reflect the latest version of the earlier classifier.

Other readings

The follow list of publicly accessible or open-access readings may be useful for further understanding of the classifiers and their differences:

• Differences between the Kraken family of classifiers: from the Kraken* authors themselves.
• de Vries, J. J. C., et al. (2021). Recommendations for the introduction of metagenomic next-generation sequencing in clinical virology, part II: bioinformatic analysis and reporting. Journal of Clinical Virology: The Official Publication of the Pan American Society for Clinical Virology, 138(104812), 104812. https://doi.org/10.1016/j.jcv.2021.104812
• de Vries, J. J. C., et al. (2021). Benchmark of thirteen bioinformatic pipelines for metagenomic virus diagnostics using datasets from clinical samples. Journal of Clinical Virology: The Official Publication of the Pan American Society for Clinical Virology, 141(104908), 104908. https://doi.org/10.1016/j.jcv.2021.104908
• Maas, M. A. M., et al. (2026). Use of metagenomics for the detection of pathogens in the environment: A scoping review. Environmental Research, 305(Pt 2), 125050. https://doi.org/10.1016/j.envres.2026.125050
• Meyer, F., et al. (2022). Critical Assessment of Metagenome Interpretation: the second round of challenges. Nature Methods, 19(4), 429–440. https://doi.org/10.1038/s41592-022-01431-4
• Sczyrba, A., et al.(2017). Critical Assessment of Metagenome Interpretation-a benchmark of metagenomics software. Nature Methods, 14(11), 1063–1071. https://doi.org/10.1038/nmeth.4458
• Seppey, M., et al. (2026). LEMMIv2: benchmarking framework for metagenomic and 16S amplicon profilers with a catalogue of evaluated tools. Genome Biology, 1–25. https://doi.org/10.1186/s13059-026-04089-9
• Seppey, M., et al. (2020). LEMMI: a continuous benchmarking platform for metagenomics classifiers. Genome Research, 30(8), 1208–1216. https://doi.org/10.1101/gr.260398.119
• Ye, S. H., et al. (2019). Benchmarking Metagenomics Tools for Taxonomic Classification. Cell, 178(4), 779–794. https://doi.org/10.1016/j.cell.2019.07.010

I get a warning during CENTRIFUGE_KREPORT or KRAKENTOOLS_COMBINEKREPORTS_CENTRIFUGE process with exit status 255

When a sample has insufficient hits for abundance estimation, the resulting report.txt file will be empty.

When trying to convert this to a kraken-style report or merging together, the tools will exit with a status code 255, and provide a WARN.

This is not an error nor a failure of the pipeline, just your sample has no hits to the provided database when using Centrifuge.

Why does any error or failed process from BRACKEN_BRACKEN get ignored?

If Kraken2 doesn’t classify any reads (100% unclassified) Bracken will fail on that input. Having no reads taxonomically classified can be a reasonable outcome for some samples, and we don’t want to treat this as an error.

Therefore to allow the remainder of a given run to complete if a single Bracken task fails, we set the Nextflow errorStrategy to ignore. You will still get a warning in the Nextflow console output and log that a Bracken task failed.

If you want to change this behaviour, you can override the errorStrategy in your own Nextflow configuration file.

For example:

process {
    withName: BRACKEN_BRACKEN {
        errorStrategy = 'retry'
    }
}

Other errorStrategy options can be found on the Nextflow documentation.