Quality Control for Long-Read Sequencing

  • Quality control (QC) for long-read sequencing is an essential part of bioinformatics because it evaluates the accuracy, length, completeness, and reliability of sequencing data before downstream analysis. Long-read technologies, particularly Oxford Nanopore Technologies (ONT) and Pacific Biosciences (PacBio), generate reads that can span thousands to hundreds of thousands of nucleotides. These long reads provide important advantages for genome assembly, structural variant detection, haplotype phasing, transcriptome analysis, and the study of repetitive genomic regions. However, their data have platform-specific characteristics, including different error profiles, read-length distributions, signal properties, and quality metrics. Therefore, long-read quality control requires approaches that consider both sequencing technology and the biological purpose of the experiment.
  • The quality of long-read sequencing data is influenced by several stages, including sample collection, DNA or RNA extraction, nucleic acid integrity, library preparation, sequencing chemistry, instrument performance, basecalling, and computational processing. Poor-quality or degraded DNA can produce shorter reads, while contamination, inefficient library preparation, and inappropriate handling can reduce sequencing yield. For RNA sequencing, RNA degradation, incomplete transcript coverage, and reverse-transcription artifacts may affect the quality and interpretation of the results. Quality control should therefore begin before sequencing and continue through raw signal processing, basecalling, read filtering, alignment or assembly, and final data evaluation.
  • One of the most important quality indicators in long-read sequencing is read length. Unlike short-read sequencing, which generally produces reads of relatively uniform length, long-read sequencing generates a broad distribution of read lengths. The distribution depends on the sequencing platform, sample quality, library preparation method, and experimental conditions. PacBio HiFi sequencing commonly produces reads of several kilobases to tens of kilobases, while Oxford Nanopore sequencing can generate reads ranging from short fragments to hundreds of kilobases or longer when high-molecular-weight DNA is successfully prepared. Read length is important because longer reads can span repetitive regions, connect distant genomic sequences, support haplotype phasing, and improve genome assembly. However, longer reads are not automatically better in every analysis. A dataset containing a small number of extremely long reads may still have insufficient total coverage or poor overall quality.
  • Read-length statistics commonly include the minimum, maximum, mean, median, and distribution of read lengths. The N50 read length is another widely used measure. It is the read length at which 50% of the total sequenced bases are contained in reads of that length or longer. A higher N50 often indicates a greater contribution from long reads, although it should not be interpreted independently of total yield, read accuracy, and coverage. The N50 statistic is particularly useful for comparing long-read datasets, but it does not describe the accuracy of individual reads or the completeness of a genome assembly.
  • Sequencing yield is another important QC parameter. It refers to the total number of bases generated during a sequencing run. Yield may be reported as the total number of bases, the number of reads, or the amount of data passing specified quality filters. A dataset can have a high total yield but still contain many short or low-quality reads. Therefore, yield should be evaluated together with read-length distribution, quality scores, and the proportion of reads retained after filtering. In genome sequencing, the total yield can be compared with the estimated genome size to calculate approximate sequencing coverage. For example, if 90 gigabases of usable sequence data are generated for an organism with a 3-gigabase genome, the theoretical average coverage is approximately 30-fold, assuming the reads represent the genome uniformly.
  • Read accuracy is a central consideration in long-read sequencing. It describes how closely a sequenced read corresponds to the original biological molecule. Errors may include substitutions, insertions, and deletions. The relative frequency of these errors depends on the sequencing platform, chemistry, basecalling model, library preparation, and data-processing method. Oxford Nanopore sequencing measures changes in ionic current as nucleic acids pass through a nanopore, and basecalling software converts these electrical signals into nucleotide sequences. PacBio sequencing detects nucleotide incorporation in real time, and circular consensus sequencing can use multiple observations of the same DNA molecule to produce highly accurate HiFi reads. Because the technologies use different measurement principles, their error patterns and quality metrics should not be assumed to be identical.
  • Phred quality scores are commonly used to express the estimated probability of an incorrect base call. A higher Phred score indicates a lower estimated probability of error. The relationship is expressed as:

[
Q = -10\log_{10}(P)
]

  • where (Q) is the Phred quality score and (P) is the estimated probability that a base call is incorrect. For example, a quality score of Q20 corresponds to an estimated error probability of 1%, while Q30 corresponds to 0.1%. Long-read FASTQ files generally contain per-base quality scores, allowing researchers to evaluate read accuracy at the base level. However, quality scores are estimates generated by basecalling systems and should not be treated as direct experimental measurements of accuracy. Their calibration can vary between platforms and basecalling models.
  • Oxford Nanopore sequencing requires particular attention to raw signal quality and basecalling. The instrument records electrical current measurements as DNA or RNA molecules move through nanopores. These measurements are stored in signal-data formats such as POD5, while older datasets may use FAST5. Basecalling software interprets the signal and converts it into nucleotide sequences with associated quality scores. Dorado is a commonly used ONT basecaller, and its models and configuration can influence the resulting read accuracy. Researchers should record the basecaller version, model, and settings because different models may produce different results from the same raw signal data. Where appropriate, they should also consider whether simplex or duplex reads are being analysed, since duplex sequencing combines information from both strands of a DNA molecule to improve sequence accuracy.
  • PacBio sequencing has its own quality-control requirements. In SMRT sequencing, DNA molecules are prepared as circular SMRTbell templates and observed during repeated sequencing passes. For suitable libraries, multiple passes over the same molecule can be combined to generate a circular consensus sequence (CCS). Reads meeting PacBio’s specified accuracy criteria are commonly described as HiFi reads. Important QC metrics include the number of polymerase reads, subreads, CCS reads, HiFi reads, read-length distribution, number of passes, and predicted read accuracy. The proportion of reads successfully converted into HiFi reads is particularly important because it indicates how effectively the sequencing data support high-accuracy consensus generation. PacBio analysis commonly uses SMRT Link and related tools for processing and quality assessment.
  • Raw signal quality, basecalling quality, and read quality are related but distinct concepts. Raw signal quality concerns the electrical or optical measurements produced by the sequencing instrument. Basecalling quality concerns the reliability of the nucleotide sequence inferred from those measurements. Read quality combines information about base-level accuracy, read length, completeness, and other characteristics. A dataset may contain high-quality raw signals but still require careful basecalling, or it may contain long reads with relatively low base-level accuracy. These different dimensions should be evaluated separately whenever possible.
  • Read-length distribution and quality-score distribution should be examined together. For example, a dataset may contain many very long reads, but if those reads have poor basecalling quality, they may not be suitable for high-confidence variant detection. Conversely, a dataset containing shorter but highly accurate reads may be useful for small-variant analysis, although it may provide less information for resolving long repetitive regions. The appropriate balance depends on the intended application. Genome assembly, structural variant detection, transcriptome analysis, and methylation analysis may require different filtering strategies.
  • Adapter and barcode contamination should also be investigated. Long-read library preparation commonly introduces adapters, motor proteins, barcodes, or other technical sequences. These elements may appear in raw reads and can interfere with alignment, assembly, or downstream interpretation if not handled correctly. Oxford Nanopore workflows may require removal of sequencing adapters and barcode sequences, while PacBio workflows may require processing of circular templates and removal or interpretation of adapter-related sequences. Tools such as Porechop have historically been used for ONT adapter trimming and demultiplexing, although compatibility and maintenance should be checked before selecting software for a current workflow. Cutadapt and other sequence-processing tools may also be useful when their supported input and adapter configurations match the dataset. Researchers should avoid removing biologically meaningful barcodes or molecular identifiers that are required for downstream analysis.
  • Contamination screening is another important component of long-read QC. Contamination may originate from host organisms, microorganisms, laboratory reagents, environmental sources, or cross-sample carryover. In metagenomic sequencing, the presence of multiple organisms is expected, so taxonomic classification must distinguish genuine community members from unwanted contaminants. In host-associated pathogen sequencing, reads originating from the host may need to be identified and removed before microbial analysis. Tools such as Kraken2, Centrifuge, and Kaiju can support taxonomic classification, while reference-based alignment using minimap2 or other suitable aligners can help identify reads matching known host or contaminant sequences. The reference database must be appropriate and sufficiently comprehensive, because incomplete databases can lead to incorrect or unresolved classifications.
  • Read filtering is commonly performed to remove sequences that do not meet predefined quality criteria. Filtering may consider minimum read length, average quality score, minimum per-base quality, ambiguous nucleotide content, and other platform-specific characteristics. In some applications, researchers may also remove reads containing unwanted adapters, contaminants, or technical artifacts. However, filtering thresholds should be selected according to the biological objective rather than applied automatically. Very aggressive filtering may remove useful biological information, particularly in datasets where shorter reads are expected or where low-quality regions contain important sequence information. The number of reads and bases retained after filtering should always be recorded.
  • For Oxford Nanopore data, NanoPlot is widely used to visualize read-length distributions, quality scores, and other sequencing metrics. NanoStat provides summary statistics, while pycoQC can be used to assess relevant run-level sequencing information from supported input data. NanoFilt can filter or trim reads based on quality and length criteria. These tools can help researchers identify poor-quality runs, unusual read-length distributions, and changes in data quality across sequencing batches. Their suitability depends on the file formats and metadata available in a particular workflow.
  • For PacBio data, SMRT Link provides tools for processing and assessing sequencing output, including CCS and HiFi read generation. PacBio’s command-line tools, including those available through the pbtk toolkit, can support additional processing and data inspection. The exact metrics depend on the sequencing mode and software version. Researchers should consult the relevant platform documentation when interpreting polymerase reads, subreads, CCS reads, HiFi yield, and predicted accuracy.
  • MultiQC can be used to aggregate quality-control reports from multiple samples and supported bioinformatics tools. It is particularly useful when many sequencing libraries are processed together because it provides a consolidated overview of quality metrics. NanoPlot and other long-read tools may generate reports that can be incorporated into a broader QC workflow, depending on the supported input formats and software versions. FastQC is primarily designed for short-read FASTQ quality assessment and does not provide a complete assessment of long-read sequencing characteristics. Therefore, long-read-specific tools should be used alongside general QC software rather than relying on FastQC alone.
  • A simple Oxford Nanopore quality-control workflow may begin with a basecalled FASTQ file. NanoStat can summarize read statistics, NanoPlot can generate visual reports, and NanoFilt can be used to retain reads that meet selected quality and length thresholds. For example, the following commands illustrate a basic workflow:

NanoStat –fastq reads.fastq.gz

NanoPlot –fastq reads.fastq.gz –outdir nanoplot_report

gunzip -c reads.fastq.gz | NanoFilt -q 10 -l 1000 | gzip > filtered_reads.fastq.gz

  • In this example, reads are retained if they meet a minimum average quality threshold of Q10 and a minimum length of 1,000 bases. These values are illustrative rather than universal recommendations. The appropriate thresholds depend on the sequencing platform, basecalling model, read distribution, and downstream analysis. Researchers should inspect the original dataset before deciding whether these filters are suitable.
  • A simple PacBio HiFi workflow may involve examining the sequencing output and generating consensus reads from subreads when CCS processing is required. The following command illustrates CCS generation from a PacBio subread BAM file:

ccs input.subreads.bam output.ccs.bam

  • The actual command may require additional options depending on the instrument, library, software version, and desired processing settings. HiFi reads can subsequently be exported or processed for downstream applications using compatible PacBio tools. Researchers should verify the resulting read count, read-length distribution, and predicted accuracy rather than assuming that all input molecules generated usable HiFi reads.
  • Quality control should also consider sequencing-run performance. Important metrics may include total yield over time, read production rate, active channels or sequencing wells, read-length distribution, quality-score distribution, and the proportion of data passing the selected filters. For Oxford Nanopore sequencing, channel activity and run-level output can help identify changes in pore performance or library loading. For PacBio sequencing, instrument and run metrics provide information about productive sequencing activity and the generation of usable consensus reads. These metrics can help distinguish problems related to sample preparation from those related to instrument performance.
  • Long-read quality control differs according to the downstream application. In genome assembly, read length, total yield, read accuracy, and coverage are especially important because they influence contiguity, repeat resolution, and consensus quality. Tools such as NanoPlot, NanoStat, and platform-specific processing software can assess raw read characteristics, while assembly evaluation tools such as QUAST and BUSCO help examine the resulting assembly. For microbial genome assemblies, CheckM2 may provide additional information about genome completeness and contamination. Assembly-level QC is complementary to raw-read QC and should not replace it.
  • For structural variant detection, long reads are valuable because they can span large insertions, deletions, inversions, duplications, and complex rearrangements. However, low-quality reads and mapping errors may produce false variant calls. Read accuracy, alignment quality, coverage, and mapping consistency should therefore be assessed before variant interpretation. Long-read aligners such as minimap2 are commonly used, while tools such as Sniffles2, cuteSV, and SVIM can support structural variant detection. The choice of software and parameters should reflect the sequencing platform and read type.
  • For haplotype phasing, long reads can connect multiple variants on the same DNA molecule and help reconstruct chromosome-specific sequences. The usefulness of phasing depends on read length, heterozygous variant density, coverage, and accuracy. Tools such as WhatsHap and LongPhase can support phasing analyses, but their performance may vary with data quality and organismal characteristics. Low coverage or incorrect variant calls can reduce phasing accuracy, so read-level and variant-level QC are important.
  • In transcriptome analysis, long-read sequencing can capture complete or near-complete RNA molecules and reveal alternative splicing, transcript isoforms, fusion transcripts, and transcript structures. PacBio Iso-Seq and Oxford Nanopore direct RNA or cDNA sequencing are commonly used approaches. Quality control should examine read length, transcript completeness, sequencing accuracy, adapter or primer contamination, and potential artifacts introduced during reverse transcription or amplification. Tools such as NanoPlot and NanoStat can summarize ONT read characteristics, while transcriptome analysis tools such as FLAIR, TALON, StringTie2, and Bambu can support downstream processing and transcript identification. Researchers should distinguish genuine biological isoforms from incomplete transcripts, sequencing artifacts, and errors in transcript reconstruction.
  • Long-read sequencing is also increasingly used for epigenetic analysis. Oxford Nanopore sequencing can detect certain DNA modifications from changes in electrical current, while PacBio sequencing can provide information about some base modifications through polymerase kinetics. Quality control in these applications must consider the accuracy of modification detection models, the quality of the underlying sequence data, and the availability of suitable controls. Modified-base calls should not be interpreted as definitive evidence without considering model performance, coverage, sequence context, and experimental validation where necessary.
  • Metagenomic long-read sequencing requires additional attention to taxonomic classification, read length, contamination, and the representation of different organisms. Long reads can improve the recovery of microbial genomes and help connect genes to their genomic context. However, uneven abundance, host contamination, extraction bias, and differences in genome composition may affect the resulting dataset. Tools such as Kraken2, Centrifuge, and Kaiju can support taxonomic profiling, while Flye and other long-read assemblers can be used for assembly. The resulting metagenome-assembled genomes should be evaluated for completeness, contamination, and taxonomic consistency using suitable tools and reference resources.
  • The following Python example illustrates how basic read-length statistics can be calculated from a FASTQ file using Biopython. It reports the number of reads, total bases, mean read length, and median read length. It is a simple summary and does not replace platform-specific quality assessment.

from Bio import SeqIO

from statistics import mean, median

fastq_file = “reads.fastq”

lengths = []

for record in SeqIO.parse(fastq_file, “fastq”):

    lengths.append(len(record.seq))

if lengths:

    print(“Number of reads:”, len(lengths))

    print(“Total bases:”, sum(lengths))

    print(“Mean read length:”, mean(lengths))

    print(“Median read length:”, median(lengths))

else:

    print(“No reads found.”)

  • A more complete QC workflow should also examine per-base quality scores, read-length distributions, sequencing yield, adapter content, contamination, and the number of reads retained after filtering. For large compressed FASTQ files, researchers may prefer specialized tools that process data efficiently without loading all sequences into memory.
  • Another important consideration is the distinction between raw-read QC and assembly-level QC. Raw-read QC evaluates the quality of the sequencing data before assembly, whereas assembly-level QC evaluates the reconstructed genome or transcriptome. An assembly may appear highly contiguous but contain errors, misassemblies, collapsed repeats, or contamination. Conversely, a dataset with relatively fragmented reads may still produce a useful assembly when coverage and read accuracy are sufficient. Tools such as QUAST, BUSCO, and CheckM2 provide complementary information, but no single metric establishes that an assembly is biologically correct.
  • Quality control should be performed both before and after read filtering. Comparing the original and filtered datasets helps determine whether filtering improved the quality of the retained reads and whether excessive data were discarded. Useful comparisons include total bases, number of reads, mean and median read lengths, N50, quality-score distributions, and the percentage of reads retained. The percentage of retained bases is particularly important because a filtering strategy may preserve a high number of reads while removing a substantial proportion of the total sequence data.
  • Reproducibility is essential in long-read sequencing analysis. Researchers should document the sequencing platform, instrument, chemistry, library preparation method, sample type, basecaller and model, software versions, command-line parameters, reference databases, and filtering thresholds. Original raw signal data and unfiltered basecalled reads should be retained whenever possible, together with checksums and metadata. Quality-control reports should be stored with the processed data so that later analyses can be traced to the original sequencing output. Workflow managers such as Snakemake and Nextflow can help organize and reproduce multi-step QC pipelines.
  • Long-read sequencing quality control is therefore a multidimensional process involving read length, read accuracy, sequencing yield, signal quality, basecalling performance, contamination, adapter content, and application-specific metrics. Oxford Nanopore and PacBio technologies require different approaches because they generate data through different measurement principles and processing workflows. Tools such as NanoPlot, NanoStat, NanoFilt, pycoQC, SMRT Link, MultiQC, and Biopython support different parts of the QC process. By evaluating raw reads carefully, applying appropriate filtering thresholds, and validating downstream assemblies or analytical results, researchers can improve the reliability of long-read sequencing studies in genomics, transcriptomics, metagenomics, structural variation analysis, and epigenomics.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *