Version Compatibility

Reference examples tested with: NanoPlot 1.42+ (NanoPack2), cramino 0.14+, chopper 0.7+, Filtlong 0.2+, seqkit 2.5+, pycoQC 2.5+.

Before using code patterns, verify installed versions match. If versions differ:

• CLI: <tool> --version then <tool> --help to confirm flags (chopper/cramino are fast-moving Rust tools)

Inputs that determine what QC is even possible - record them:

• sequencing_summary.txt is produced by the basecaller (Dorado/Guppy), not the FASTQ. pycoQC/toulligQC REQUIRE it for pore activity, yield-over-time, and translocation speed. FASTQ-only hand-off permanently loses the run-health layer.
• Percent identity requires a reference BAM (NanoPlot --bam / cramino); it cannot come from FASTQ.

If code throws an error, introspect the installed tool (NanoPlot --help, cramino --help) and adapt the example to the actual API rather than retrying.

Long-Read QC

"Is my long-read run any good?" -> Read length N50 and yield from FASTQ, real percent identity from a reference BAM, run-health from the sequencing_summary, then filter for the downstream goal.

• CLI: NanoPlot --fastq reads.fq.gz -o qc/ (overview), cramino aln.bam (fast BAM stats + identity), pycoQC -f sequencing_summary.txt -o run.html (run health)

The Single Most Important Modern Insight -- Read-Only Qscore Is a Self-Graded Posterior; Real Accuracy and the Failures That Sink a Run Are Only Visible Against a BAM and the Summary

Three corrections a naive long-read QC misses:

1. Per-read Qscore is an uncalibrated basecaller posterior, not an empirical error rate. It is the Phred of the mean per-base error probability (NOT the arithmetic mean of Q values), assigned by the basecaller to its own output. ONT's own data: bases labeled Q20 are empirically ~Q12.5 on older chemistries; R10 sup and HiFi are better calibrated but read-only Q still overstates accuracy. Real accuracy is gap-compressed identity from a reference BAM (cramino, NanoPlot --bam). Treat Q thresholds as relative knobs, not accuracy guarantees.
2. The sequencing_summary.txt is the run-health layer, and it is not in the FASTQ. Pore/channel activity, yield-over-time, translocation speed, and barcode breakdown come from the basecaller's summary TSV. Hand a collaborator only FASTQ and that layer is gone (re-basecalling from POD5 can regenerate it; FASTQ cannot).
3. The right filter depends on intent, not a fixed cutoff. Assembly wants the long reads (which are the lowest-Q) and small replicons preserved - subsample by quality, never hard-length-cut. Variant calling wants depth - filter almost nothing and let the caller model per-base Q. HiFi is already Q20+ - do not Phred-filter it like noisy CLR.

Tool Roles

Read N50 = the length where 50% of total bases are in reads at least that long (length-weighted, far above the median); it predicts assembly contiguity. NanoFilt and the rrwick Porechop are deprecated/unmaintained (use chopper and Porechop_ABI).

Intent-Conditioned Filtering Decision Tree

Core Commands

# Overview from FASTQ (length + posterior quality only - not real accuracy)
NanoPlot --fastq reads.fq.gz -o qc_fastq/ --N50
seqkit stats -a reads.fq.gz                     # N50 + quartiles, fast

# Real accuracy: fast BAM stats incl. gap-compressed identity (needs a reference BAM)
cramino aln.bam
NanoPlot --bam aln.bam -o qc_bam/               # percent identity scatter

# Run health (requires the basecaller's summary)
pycoQC -f sequencing_summary.txt -o run_qc.html

# Compare barcodes / runs
NanoComp --bam s1.bam s2.bam s3.bam --names s1 s2 s3 -o compare/

# Filter for VARIANT calling: light quality only
chopper -q 10 -i reads.fq.gz | gzip > q10.fq.gz

# Subsample for ASSEMBLY: by quality to ~100x of a 5 Mb genome (never a hard length cut)
filtlong --target_bases 500000000 reads.fq.gz | gzip > subsampled.fq.gz

Per-Method Failure Modes

Trusting FASTQ Qscore as accuracy

Trigger: judging a run from NanoStat --fastq mean Q. Mechanism: Q is an uncalibrated posterior. Symptom: "Q20 reads" that are ~94% accurate. Fix: align and read gap-compressed identity (cramino / NanoPlot --bam).

QC without the summary

Trigger: only FASTQ/BAM at hand-off. Mechanism: run-health metrics live in sequencing_summary.txt. Symptom: cannot see pore death, mux map, or yield-over-time. Fix: obtain the summary (or re-basecall from POD5 to regenerate it).

Over-filtering erases assembly value

Trigger: a blunt -q 15 or hard 10 kb length cut before assembly. Mechanism: the longest reads are the lowest-Q; small plasmids fall under a length floor. Symptom: worse N50; missing plasmids. Fix: subsample by quality (Filtlong --target_bases), keep the long tail, never length-floor above the smallest replicon.

Chimeras masquerade as SVs

Trigger: undetected internal adapters (two molecules ligated as one read). Mechanism: the read's halves map to different loci. Symptom: phantom translocations/insertions in the SV VCF. Fix: check whether Dorado already trimmed/split; use Porechop_ABI for unknown adapters; suspect a biologically implausible long-read spike.

Re-filtering HiFi like CLR

Trigger: Phred-quality-filtering PacBio HiFi. Mechanism: HiFi is Q20+ consensus already. Symptom: wasted reads, no accuracy gain. Fix: filter on rq >= 0.99 only.

Quantitative Thresholds

Common Errors

References

• De Coster W, D'Hert S, Schultz DT, Cruts M, Van Broeckhoven C. 2018. NanoPack: visualizing and processing long-read sequencing data. Bioinformatics 34(15):2666-2669.
• De Coster W, Rademakers R. 2023. NanoPack2: population-scale evaluation of long-read sequencing data (cramino, chopper). Bioinformatics 39(5):btad311.
• Leger A, Leonardi T. 2019. pycoQC, interactive quality control for Oxford Nanopore Sequencing. J Open Source Softw 4(34):1236.
• Steinig E, Coin L. 2022. Nanoq: ultra-fast quality control for nanopore reads. J Open Source Softw 7(69):2991.
• Bonenfant Q, Noé L, Touzet H. 2023. Porechop_ABI: discovering unknown adapters in Oxford Nanopore sequencing reads. Bioinform Adv 3(1):vbac085.
• Shen W, Le S, Li Y, Hu F. 2016. SeqKit: a cross-platform and ultrafast toolkit for FASTA/Q file manipulation. PLoS ONE 11(10):e0163962.

Related Skills

• basecalling - Produces the reads and the sequencing_summary.txt this QC needs
• long-read-alignment - Produces the BAM required for real percent identity
• structural-variants - Chimeras flagged here fabricate SVs there
• medaka-polishing - QC/subsample reads before polishing
• genome-assembly/long-read-assembly - Subsample by quality before assembling
• genome-assembly/genome-profiling - K-mer ploidy/size estimate alongside read QC
• read-qc/quality-reports - General (short-read-oriented) read QC
• sequence-io/sequence-statistics - FASTA/FASTQ summary statistics