Pre-assembled contigs mode

mode: contigs — the input is a genome somebody already assembled. There is no read QC and no assembler, so the whole front end is one rule, whose job is to hand the contamination screen a FASTA it can work with.

Use it for a genome from a collaborator, from a database, or from an earlier assembly. Every analysis stage still runs; the mode cannot re-examine evidence that only exists in reads.

flowchart LR
    IN([sample.fasta<br/>a genome somebody already assembled])
    FC[filter_contigs<br/>the whole front end, one rule]
    SC[contamination screen<br/>BLAST, BlobTools, genus selector]
    OUT([contigs_final.fasta])
    IN --> FC --> SC --> OUT

Input files are {sample}.<fasta|fa|fna> in input.contigs_dir, and every file in one batch must share the extension. A dotted name such as my.genome.fasta is read as sample my.genome: the last dot splits off the extension, earlier dots stay in the name.

Terms used on this page

QC	quality control
CARD	Comprehensive Antibiotic Resistance Database
IS	insertion sequence, the smallest kind of transposable element
AMR	antimicrobial resistance

One rule, two branches

filter_contigs looks at the first header and takes one of two paths. SPAdes writes the same style for every record, so one line is enough to decide.

Branch	Trigger	What happens
A	the header matches length_<n> or cov_<n>	contigs under 500 bp or 2× coverage are dropped, both numbers read out of the header. Headers are kept exactly as they are
B	anything else	each header is trimmed to its first whitespace token. No length filter and no coverage filter. Nothing else changes

The rule logs which branch it took, in the first line of logs/filter_contigs_{sample}.log. It decides whether any filtering happened at all.

Why branch B does not invent a filter

The length and coverage of each contig are read out of the SPAdes header. A genome from NCBI, from Flye, or from another assembler carries no such numbers, and guessing them would silently delete real sequence. So branch B removes nothing, and the only contigs that can leave the assembly are the ones the contamination screen drops.

The rule name is misleading in that branch: it normalises headers rather than filtering. The normalisation matters — Bakta, Platon, geNomad and the BLAST screen all join on that first token.

Decontamination does most of the work, with one hand tied

As in illumina mode, the screen's output is contigs_final.fasta. But BlobTools combines two signals — BLAST taxonomy and read coverage — and this mode has no reads. The contigs are mapped against themselves with minimap2 purely so BlobTools has a BAM to read, which gives a near-uniform depth that carries no information.

So only the BLAST taxonomy leg is doing real work here. Every keep-or-drop decision still lands in contaminants/contig_taxonomy_decisions.tsv with its reason, and reading that file matters more in this mode than in any other. See Decontamination.

What is absent, and why

Missing	Reason
01.reads/	there are no reads to QC
05.amr/mapping/ — the CARD read screen	it maps reads. ABRicate on the contigs still runs
Qualimap	the only alignment is the contigs against themselves; a report on it would say nothing
Circular-replicon information for Bakta	that comes from Flye's topology calls. Every sequence is annotated as linear
IS copy number in the mobilome module	it is a read-mapping leg

Everything else — taxonomy, annotation, AMR and virulence screening, plasmids, prophages, the optional mobilome module and the report — runs exactly as in the other three modes.

What it writes

Path	Contents
02.assembly/{sample}/contigs_filt.fasta	branch A: the filtered assembly. Branch B: the same sequence with trimmed headers
02.assembly/{sample}/contaminants/	BLAST table, BlobTools table, the decision audit
02.assembly/{sample}/contigs_final.fasta	the delivered genome

This mode replaces the separate FastaFlux workflow of v1. If you are migrating from it, see Coming from v1.