Long-Read Sequencing: choosing a workflow

Long reads help when the biological question depends on sequence context across repeats, rearrangements, or multiple exons in one RNA molecule. Choose the platform and analysis route from the input molecules and question; “long read” alone does not specify the right tool or guarantee that a feature will resolve.

Start with the molecule and question

For genomic DNA, decide whether you need a de novo assembly, read-to-reference structural-variant analysis, or small-variant analysis. For long RNA, decide whether you need transcript structure discovery, transcript abundance, or both. These are different workflows even when they start from reads produced on the same instrument.

PacBio HiFi reads are circular-consensus reads designed to combine long read length with high per-read accuracy. Oxford Nanopore reads are generated from electrical signals and can be produced in real time; read lengths and accuracy depend on sample preparation, chemistry, basecaller and read mode. Platform labels do not replace checking the actual FASTQ/BAM metadata and run documentation.

Inspect read quality before analysis

NanoPlot summarizes long-read length and quality distributions from FASTQ and can also use alignment or sequencing-summary inputs. It is a descriptive QC tool, not a pass/fail rule. Compare samples and batches, check for unexpectedly short reads or quality shifts, and interpret the plots in light of the basecaller and library protocol. Do not apply one universal length or Q-score cutoff across platforms and assays.

Save the following as check_long_read_qc.sh and run bash check_long_read_qc.sh from your project directory. Replace the input path and choose threads for your machine. NanoPlot must already be installed. The stage requires a fresh output directory so old reports cannot be mistaken for a successful new run; choose a new directory name when rerunning instead of deleting evidence.

set -euo pipefail
input="reads.fastq.gz"
outdir="nanoplot_qc"

if [[ ! -s "$input" ]]; then
  echo "Missing or empty input: $input" >&2
  exit 1
fi
gzip -t "$input"
if [[ -e "$outdir" ]]; then
  echo "Output path already exists: $outdir; choose a fresh directory." >&2
  exit 1
fi

NanoPlot --threads 4 --fastq "$input" --outdir "$outdir"
test -s "$outdir/NanoPlot-report.html"
test -s "$outdir/NanoStats.txt"
find "$outdir" -maxdepth 1 -type f -print
echo "NanoPlot completed and both expected reports are non-empty."

Success means the command exits without error and the output directory contains report files, including NanoPlot's HTML summary and statistics file; open the HTML report and compare the read-length and quality distributions with the assay and run metadata. If the input check fails, correct the path first. If NanoPlot fails or produces no report, stop and check that the file is readable FASTQ, its compression is valid, and the installed NanoPlot version's --help output matches the invocation. A good-looking QC plot does not establish mapping accuracy or variant validity.

The report names above are NanoPlot's defaults without --prefix; adapt the assertions if you deliberately change output naming. Shell checks covered missing input, failed commands, incomplete reports and stale output directories using mock tool responses. NanoPlot itself and a real long-read dataset were not executed for these checks.

Genome assembly and polishing

For a bacterial de novo assembly from raw Oxford Nanopore reads, Flye documents the --nano-raw mode. The command below illustrates the interface, not a complete project recipe: real analysis also needs an appropriate input dataset, resource allocation, version tracking, and assembly evaluation. The output is a set of contigs and run files, not proof of a complete or correct genome.

# Example only: raw ONT reads, with output written to a new directory
flye --nano-raw reads.fastq.gz --out-dir flye_assembly --threads 4

Use a mode matching the read technology and accuracy profile; do not pass HiFi reads as raw ONT data. Review Flye's log and assembly statistics, then assess completeness, contamination and structural consistency using methods appropriate to the organism and study. Polishing choices depend on the platform, basecaller and available reads. Medaka is ONT-specific; Racon is a separate polishing option, not an automatic requirement or a guarantee that residual errors are removed.

Long-read RNA: alignment and isoform analysis

Long RNA reads can preserve exon linkage across an individual observed molecule, which helps distinguish splice patterns. The reads are still subject to truncation, sequencing/basecalling errors, uneven transcript representation and alignment ambiguity. A read supporting a splice pattern is evidence for that molecule, not by itself proof of a stable, expressed biological isoform or its prevalence.

For long spliced reads, minimap2 documents the splice preset. SAM output requires -a; the mapper writes SAM to standard output, so shell redirection saves it. The reference must be the genome build expected by downstream analysis.

# Example: long-read RNA alignment to a genome reference
minimap2 -ax splice -t 4 reference.fa rna_reads.fastq.gz > aligned.sam

This creates a SAM file; it does not sort or index it. For IsoQuant's aligned-read input, provide sorted and indexed BAM files, plus the matching reference and (when using annotation) gene annotation. The original page's command used isoquant.py and supplied BAM without showing those preparation steps. Current upstream quick-start uses isoquant; check the installed release's documented CLI before adapting a command.

# Illustrative IsoQuant invocation; aligned.bam must already be sorted and indexed
isoquant --reference reference.fa \
  --genedb annotation.gtf \
  --bam aligned.bam \
  --data_type nanopore \
  -o isoquant_results

Set --data_type to match the reads (for example, nanopore or pacbio_ccs), and confirm details against the installed release. Keep poly(A) tails when following IsoQuant's recommendations for transcript model construction. Inspect logs and output tables; compare inferred transcript structures with annotation and independent evidence before calling a transcript novel.

Record an analysis plan

1. Define the biological question. State whether the primary output is an assembly, structural-variant evidence, or transcript/isoform catalogue and quantification.
2. Record the input contract. Note DNA or RNA, platform and read mode (for example HiFi, ONT cDNA, or direct RNA), basecaller where relevant, file type, and any preprocessing.
3. Check reference compatibility. Record genome assembly accession/build and annotation release. Genome and annotation must describe the same coordinate system.
4. Review QC and alignment. Save read-length/quality summaries, mapping summaries, and failed-read/filtering decisions. Report metrics rather than silently applying a universal threshold.
5. Validate the interpretation. For assemblies, inspect contiguity and completeness indicators and unresolved regions. For variants or isoforms, inspect supporting reads and use orthogonal or independent evidence appropriate to the claim.

Further reading

• NanoPlot upstream documentation
• Flye repository and usage manual
• minimap2 manual
• IsoQuant upstream documentation
• PacBio: how HiFi sequencing works
• Oxford Nanopore documentation repository