← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-genome-assembly-assembly-qc / references / usage-guide.md

Assembly QC - Usage Guide

Overview

Assembly QC judges whether a genome assembly is good enough to annotate, submit, or publish by measuring three orthogonal axes: contiguity (how fragmented), completeness (is everything there), and correctness (is what is there right). The central lesson is that no single number is quality - least of all N50, the most-gamed metric in genomics. The modern, reviewer-expected standard reports reference-free metrics: auN/NGx for contiguity, BUSCO or compleasm gene-space completeness plus Merqury k-mer completeness, and a Merqury QV for per-base accuracy, with reference-based QUAST reserved for the special case of a same-organism reference. The agent picks methods by whether a trusted reference exists, the read chemistry, the ploidy, and the downstream goal.

Prerequisites

conda install -c bioconda quast busco compleasm merqury meryl minimap2 inspector merfin genomescope2
# k8 + calN50.js/paftools.js ship with minimap2; CRAQ installs from its GitHub repo
# BUSCO/compleasm lineage datasets and the GenomeScope2 model download separately

Large downloads to budget for: BUSCO/compleasm lineage datasets (use --offline + a local path on HPC), and a meryl k-mer database built from the raw reads (can be large for big genomes).

Quick Start

Tell your AI agent what you want to do: - "Assess the quality of my genome assembly across contiguity, completeness, and correctness" - "Compute a reference-free QV for my assembly with Merqury" - "Run BUSCO and compleasm on my assembly and tell me which to trust" - "My BUSCO Duplicated is high - is that haplotigs or a real whole-genome duplication?" - "Compute auN and NG50 instead of plain N50"

Example Prompts

Reference-free QC of a novel genome

"I assembled a non-model insect from HiFi reads and have no trusted reference. Give me a reference-free QC: Merqury QV and k-mer completeness, compleasm completeness, and auN/NG50, and tell me whether it clears the EBP 6.C.Q40 bar."

"Run Inspector on my Nanopore assembly to find structural errors without a reference, and explain why I should not use QUAST against the congener genome I have."

Diagnosing a duplicated or fragmented assembly

"My assembly is about 1.8x the GenomeScope2 size estimate and BUSCO Duplicated is 14%. Walk through whether this is uncollapsed haplotigs or a real WGD, and what to check before running purge_dups."

"BUSCO Complete is 91% on my T2T-grade HiFi assembly, which seems low. Re-run completeness with compleasm and explain whether the gap is the assembly or BUSCO's predictor."

QV and the circularity trap

"I polished my assembly with Illumina reads. Build a Merqury k-mer database and compute QV, and make sure the QV is not measured on the same reads I polished with."

Comparing assemblers

"Compare my SPAdes, Flye, and hifiasm assemblies on contiguity (auN/NG50), completeness (compleasm), and accuracy (Merqury QV), and tell me which is best and why N50 alone would mislead me."

Phased diploid QC

"I have a trio-phased hifiasm assembly with parental reads. Compute switch and hamming error with Merqury hap-mers and interpret the hap-mer blob plot."

What the Agent Will Do

  1. Establish whether a trusted same-organism reference and a genome-size estimate exist, and choose reference-free methods by default.
  2. Compute contiguity with auN/NGx (calN50/paftools) and QUAST, reporting contig and scaffold N50 separately.
  3. Score gene-space completeness with BUSCO and/or compleasm at the deepest applicable clade lineage, recording the OrthoDB generation.
  4. Build a meryl k-mer database from accurate reads at the best_k.sh-derived k and run Merqury for QV, k-mer completeness, and the spectra-cn copy-number plot.
  5. For novel genomes, map raw reads back with Inspector or CRAQ for reference-free structural errors; for same-organism references, add QUAST structural metrics.
  6. For phased assemblies, compute switch/hamming error and read the hap-mer blob plot.
  7. Triangulate high BUSCO-Duplicated against size, spectra-cn, and asmgene to call haplotigs vs WGD, and recommend purge_dups or not.
  8. Report all three axes with method and read-source, compare against the EBP/VGP bar appropriate to the organism, and flag any axis that is missing.

Tips