bio-assembly-qc

Bio Assembly QC

Assemble isolate genomes or metagenomes from QC-passed reads, normalize the assembler output to one contigs.fasta per sample, and run QUAST or MetaQUAST.

Instructions

Tool guides and versions: docs/README.md.

1. Validate the assembly manifest and inspect a restartable execution plan before starting expensive work:

   uv run --script skills/bio-assembly-qc/scripts/run_assembly_qc.py \
     assemblies.tsv --out results/bio-assembly-qc --threads 4
   # Inspect run_manifest.json, then execute the same plan on a compute node:
   uv run --script skills/bio-assembly-qc/scripts/run_assembly_qc.py \
     assemblies.tsv --out results/bio-assembly-qc --threads 16 --execute
   

   The driver rejects samples whose upstream read_qc_status is not passed. It runs SPAdes (--isolate), metaSPAdes, Flye/metaFlye, or metaMDBG by mode, passes --threads to every tool (default 4; SPAdes otherwise uses 16), normalizes the assembler output to per-sample contigs.fasta, and chooses QUAST or MetaQUAST from the mode. It reuses a stage only when its declared outputs are non-empty and its .done marker exists. Run --execute through the scheduler (sbatch) for anything beyond the fixtures, not on a login node.

2. Select an assembler based on read type, genome/metagenome scope, and sample diversity:

   • Illumina short-read isolates and hybrid assemblies: SPAdes v4.0.0+ (v4.3.0 current). Use metaSPAdes for short-read metagenomes.
   • Long-read bacterial isolates (PacBio CLR, ONT): Flye v2.9.5+ for the draft assembly. Use Autocycler v0.6+ when a complete bacterial consensus genome is needed from several independent long-read assemblies; do not use it for mixed communities.
   • Long-read metagenomes: Flye v2.9.5+ in --meta mode (metaFlye) as the baseline for ONT and CLR communities.
   • HiFi metagenomes: metaMDBG v1.1+. In its benchmark it recovered up to twice as many circularized high-quality MAGs as metaFlye on HiFi data (see docs/README.md). Keep metaFlye as a comparator when a per-sample failure is suspected.
   • Very large or diverse long-read datasets where runtime limits the work: myloasm, when its read profile matches the dataset.
   • The driver does not run Autocycler, hybrid SPAdes, or myloasm. Run those by hand and record the exact commands and versions.

3. Run assembly with resource-aware settings and record exact CLI, version, thread count, and RAM ceiling.

   • For very large ONT/metagenome FASTQs, use /bio-reads-qc-mapping guidance for filtering and avoid redundant full-file raw-read preflights before filtering. Record raw file metadata (stat path, size, mtime), optionally run a small sampled check, and write seqkit stats after each produced read set.
   • Use atomic output patterns for long-running filters and assemblies: write to .tmp, verify non-empty/readable output, then mv into the final path. Resume mode should skip existing final outputs only after sanity checks; when checks fail, use a tool-supported overwrite option or remove the corrupt final output before rerunning.
   • For Flye/metaFlye failures or interrupted jobs, prefer --resume or --resume-from in the existing output directory when the prior run is structurally intact. Do not delete a large partial assembly unless logs or missing stage files show it is corrupted.

4. Summarize the QUAST v5.3+ or MetaQUAST report.tsv per sample (contig count, total length, N50, largest contig) against project thresholds.

5. For every produced contigs.fasta, invoke /tracking-taxonomy-updates to run the BBTools-container QuickClade percontig domain screen before choosing downstream genome/MAG/viral/eukaryotic workflows.

6. Use the QuickClade domain routing table to decide the next step:

   • Bacteria/Archaea -> /bio-gene-calling, /bio-annotation, and GTDB-Tk taxonomy assignment.
   • Viral or virus-like -> /bio-viromics before prokaryotic MAG tooling.
   • Eukaryota -> eukaryote-aware gene/QC workflows and EukCC where bins or genomes are present.
   • Mixed/low-confidence -> split or flag contigs before domain-specific analysis.