Workflow architecture

BacFlux is one workflow with four ways in. Whichever way you enter, you leave through the same nine numbered stages, and the mode key in the config decides nothing except how a set of contigs gets produced in the first place.

Terms used on this page

ONT	Oxford Nanopore Technologies — the long-read sequencing platform
AMR	antimicrobial resistance
EFSA	European Food Safety Authority, whose thresholds the ABRicate leg applies

The shape of a run

The diagram is in two labelled halves. Everything in the first box changes with the mode you chose; everything in the second box is the same whichever mode ran. That is why there is one workflow instead of four.

flowchart TD
    subgraph P1["PART 1   what changes with the mode you chose"]
        direction TB
        I[illumina<br/>paired short reads]
        N[nanopore<br/>ONT long reads]
        H[hybrid<br/>short reads and long reads]
        C[contigs<br/>an assembly you already have]

        IF[fastp trimming, PhiX removal<br/>assembly with SPAdes]
        NF[Filtlong, assembly with Flye<br/>Medaka polishing, dnaapler]
        HF[both front ends, then Polypolish<br/>and a Snippy comparison]
        CF[length filter only]

        I --> IF
        N --> NF
        H --> HF
        C --> CF

        DRAFT([a draft assembly])
        IF --> DRAFT
        NF --> DRAFT
        HF --> DRAFT
        CF --> DRAFT
    end

    subgraph P2["PART 2 &nbsp; the same whichever mode ran"]
        direction TB
        DECON[02.assembly<br/>decontamination: BLAST, BlobTools, genus selector]
        FINAL([contigs_final.fasta<br/>every stage below reads this one file])
        DECON --> FINAL

        TAX[03.taxonomy<br/>GTDB-Tk]
        ANN[04.annotation<br/>Bakta, eggNOG, antiSMASH, dbCAN]
        AMR[05.amr<br/>ABRicate, and BBMap against CARD]
        PLA[06.plasmids<br/>Platon, optionally geNomad]
        PHA[07.phages<br/>VirSorter2 or geNomad, then CheckV]
        REP[09.report<br/>MultiQC]

        FINAL --> TAX
        FINAL --> ANN
        FINAL --> AMR
        FINAL --> PLA
        FINAL --> PHA

        MOB[08.mobilome<br/>OFF unless you switch it on<br/>ISEScan, CONJScan, AMRFinderPlus]
        LADDER([mobility table<br/>one row per AMR gene, tier 1 to 6])

        AMR --> MOB
        PLA --> MOB
        MOB --> LADDER

        TAX --> REP
        ANN --> REP
        PHA --> REP
    end

    DRAFT --> DECON

Only 08.mobilome is optional. Everything else runs on every sample, in every mode.

Why the entry points converge so early

The four modes exist only because a genome can arrive as reads or as contigs, and reads can be short, long or both. Once there is a decontaminated assembly, nothing downstream cares how it was made — which is why Part 1 is short and Part 2 is everything else. The four v1 workflows shared roughly two-thirds of their code and had to be kept in step by hand; here that two-thirds exists once.

One thing does stay mode-aware: a short-read assembly is more fragmented, so every mobilome call carries contig-edge flags and anything spanning contigs is capped at low confidence — see the mobilome overview.

What runs where

stage	runs in	optional?
01.reads	every mode that has reads	—
02.assembly	all four	—
03.taxonomy	all four	—
04.annotation	all four	—
05.amr	all four; the CARD read leg needs short reads	—
06.plasmids	all four	Platon always; the geNomad second opinion is opt-in
07.phages	all four	always runs; VirSorter2 by default, geNomad optional
08.mobilome	all four	off by default
09.report	all four	—

The three AMR legs, and why there are three

The one place BacFlux deliberately does the same job three times. Each leg fails in a different way, so the three together say more than any one of them.

flowchart LR
    R([trimmed reads])
    F([contigs_final.fasta])

    A[ABRicate on the contigs<br/>8 databases, EFSA 80% and 70%]
    B[BBMap, reads against CARD<br/>at least 70% of the gene covered]
    P[AMRFinderPlus on the proteins<br/>curated cutoffs, one per gene]

    F --> A
    R --> B
    F --> P

    A --> AW[blind spot: a gene the<br/>assembler collapsed]
    B --> BW[blind spot: no gene boundaries,<br/>and it needs short reads]
    P --> PW[blind spot: only with the<br/>mobilome module switched on]

    A --> S([AMR summary matrix])
    B --> CR([CARD report, grouped by<br/>CARD's own resistance mechanism])
    P --> M([mobility ladder, tier 1 to 6])

Read the detail on the AMR page.

Where the files land

output_dir/
├── 01.reads/       trimmed, PhiX-free, filtlong-filtered reads   (temporary)
├── 02.assembly/    draft and decontaminated contigs, CheckM, QUAST
├── 03.taxonomy/    GTDB-Tk classification
├── 04.annotation/  Bakta, eggNOG, antiSMASH, dbCAN
├── 05.amr/         ABRicate tables, CARD read mapping
├── 06.plasmids/    Platon calls, optional geNomad concordance
├── 07.phages/      prophage predictions, CheckV quality
├── 08.mobilome/    mobility table                    (only when switched on)
├── 09.report/      MultiQC
└── logs/           one log per rule, per sample

Every path and every file is listed on the output reference page.