Long-Read Quality Control - Usage Guide

Overview

Long-read QC assesses Oxford Nanopore and PacBio run quality and filters reads for the downstream goal. It differs fundamentally from short-read QC: there is no per-cycle quality plot, the headline metric is read N50 (length-weighted), and the reported per-read Qscore is an uncalibrated basecaller posterior - real accuracy (percent identity) requires aligning to a reference. Run-health metrics (pore activity, yield-over-time, translocation speed) come from the basecaller's sequencing_summary.txt, not the FASTQ. The right filter is intent-dependent: assembly preserves the long, low-quality reads and small replicons; variant calling filters almost nothing; HiFi is already Q20+ and is filtered on its rq tag. This skill covers NanoPlot, cramino, NanoComp, pycoQC/toulligQC, seqkit, chopper, and Filtlong, plus the chimera trap that fabricates SVs and the run-health red flags an expert reads.

Prerequisites

conda install -c bioconda nanoplot nanocomp cramino chopper filtlong seqkit pycoqc
# toulligQC, fastcat, Porechop_ABI optionally for ONT-native QC and adapter discovery
# Keep the basecaller's sequencing_summary.txt for run-health QC

Quick Start

Tell your AI agent what you want to do: - "Run NanoPlot QC on my Nanopore reads and report the read N50" - "Get the real percent identity of my reads against the reference" - "Check my run health from the sequencing summary" - "Filter my reads appropriately for assembly"

Example Prompts

Run overview and real accuracy

"Give me an overview QC of my ONT FASTQ (length N50, yield), then align to the reference and tell me the real gap-compressed identity - I know the FASTQ Qscore overstates accuracy."

Run health

"From my sequencing_summary.txt, tell me whether the run had pore death, translocation-speed drift, or a high unclassified-barcode fraction."

Intent-conditioned filtering

"Filter my deep Nanopore data for a bacterial assembly. Subsample by quality to about 100x and make sure you do not erase small plasmids with a length cut."

Comparing barcodes

"Compare read length, quality, and identity across my four barcodes and flag any outlier."

Chimera check

"My SV calls have suspicious translocations. Check whether chimeric reads (internal adapters) could be the cause."

What the Agent Will Do

Summarize length distribution, read N50, and yield from the FASTQ (NanoPlot/seqkit).
Align (or use an existing BAM) and report gap-compressed identity (cramino / NanoPlot --bam) as the real accuracy.
Read run-health from the sequencing_summary.txt (pycoQC/toulligQC): pore activity, yield-over-time, translocation speed, barcode breakdown.
Choose a filter conditioned on the downstream goal (assembly vs variant calling vs HiFi vs cDNA).
For assembly, subsample by quality with Filtlong rather than a hard length cut.
Flag run-health red flags and possible chimeras.

Tips

The FASTQ Qscore is an uncalibrated posterior; align to a reference for real percent identity.
Always keep the sequencing_summary.txt; pycoQC/toulligQC need it and FASTQ-only hand-off loses run-health forever.
For assembly, subsample by quality (Filtlong --target_bases) and never apply a hard length floor above your smallest replicon - the longest reads are the lowest-Q and small plasmids vanish.
For variant calling, filter almost nothing; the caller models per-base Q and wants depth.
HiFi is Q20+ already; filter on rq >= 0.99, not on Phred quality.
Chimeras (internal adapters) masquerade as SVs; check whether Dorado already trimmed/split, and use Porechop_ABI for unknown adapters.
NanoFilt and the rrwick Porechop are deprecated; use chopper and Porechop_ABI.

basecalling - Produces reads and the sequencing_summary.txt
long-read-alignment - Produces the BAM for real percent identity
structural-variants - Chimeras flagged here fabricate SVs
genome-assembly/long-read-assembly - Subsample by quality before assembling
read-qc/quality-reports - General short-read-oriented QC
sequence-io/sequence-statistics - FASTA/FASTQ summary statistics