SWAM-meta

Snakemake workflow for end-to-end metagenomic analysis of antibiotic resistance in environmental bacterial communities.

What it produces

SWAM-meta starts from paired-end FASTQs and generates:

• fastp_summary.csv - per-sample read QC and Nonpareil coverage
• assembly_qa.tsv - per-sample assembly size, N50, and read-mapping stats
• contig_summary.tsv - per-contig abundance, taxonomy, AMR, MGE, and molecule type
• AMR_abundance_summary.csv - per-sample AMR abundance plus additive and multiplicative risk scores
• mag_summary.tsv - per-MAG abundance and annotations

It also keeps lower-level outputs under out_dir/data/, including assemblies, geNomad results, MobMess output, MAG files, and the short-read tables data/QAQC/short_reads_output.csv and data/QAQC/markers_cpg.csv.

---

Pipeline at a glance

1. Short reads - fastp QC, host filtering, KMA AMR alignment, SCG alignment, anthropogenic marker alignment, Nonpareil
2. Contigs - MEGAHIT assembly, geNomad classification, MobMess, Prodigal, AMRFinderPlus, MobileElementFinder, MMseqs2 taxonomy, contig abundance
3. MAGs - MetaBAT2 binning, per-bin AMR/MGE annotation, abundance, and optional GTDB-tk, CheckM2, and METABOLIC
4. Summary - sample-level AMR abundance and risk scoring from AMR, mobility, host context, and marker-derived exposure

---

Installation

git clone https://github.com/<org>/SWAM-meta.git
cd SWAM-meta
conda install -c conda-forge -c bioconda "snakemake>=8" snakemake-executor-plugin-slurm

All workflow tools are installed automatically into per-rule conda environments on first run with --use-conda.

Always use --scheduler greedy. The default MILP scheduler requires the cbc solver, which is not installed.

---

Quick start

Test mode

This runs the full workflow on two bundled mock samples and mini databases.

snakemake -n --use-conda --cores 4 --scheduler greedy --config test=True
snakemake    --use-conda --cores 4 --scheduler greedy --config test=True

Expected runtime is about 30-60 minutes on a laptop. Outputs go to test/output/.

Production mode

Edit config/config.yaml:

in_dir:  /path/to/fastq_files
out_dir: /path/to/output

Then run:

snakemake -n --use-conda --scheduler greedy
snakemake    --use-conda --cores <N> --scheduler greedy

To resume after a failed run:

snakemake --use-conda --cores <N> --scheduler greedy --rerun-incomplete

---

Databases

Most databases are downloaded and prepared automatically in dbs/ on first use, including:

• AMRFinderPlus databases
• human reference for host filtering
• anthropogenic markers (pBI143, crAss001)
• geNomad database
• UniRef50 MMseqs2 taxonomy database
• CheckM2 database
• METABOLIC

GTDB-tk is the only manual setup step.

conda run -n gtdbtk download-db.sh /path/to/gtdbtk_db

Then set:

gtdbtk_db: /path/to/gtdbtk_db

Optional stages can be disabled in config/config.yaml:

skip_gtdbtk: False
skip_metabolic: False
skip_checkm2: False
genomad_splits: 1

Increase genomad_splits if geNomad needs less peak memory.

---

Running on SLURM

Two profiles are included:

Profile	Best for
config/slurm/small-batch	Fewer than 50 samples
config/slurm/large-batch	50 or more samples

Set your account and partition in each profile, then run one of:

snakemake --profile config/slurm/small-batch
snakemake --profile config/slurm/large-batch

The profiles already enable use-conda and the required greedy scheduler.

---

Running only part of the workflow

Use these config flags:

run_short_reads: True
run_contigs: True
run_mags: True

run_short_reads	run_contigs	run_mags	Result	Extra input needed
True	True	True	Full pipeline	None
True	True	False	Short reads + contigs	None
True	False	False	Short reads only	None
False	True	True	Contigs + MAGs	clean_reads_dir
False	True	False	Contigs only	clean_reads_dir
False	False	True	MAGs only	clean_reads_dir, contigs_dir

When skipping upstream stages:

• clean_reads_dir must contain host-filtered paired FASTQs
• contigs_dir must contain {sample}.contigs.fa

If run_short_reads: False, contig abundance falls back to mean depth rather than fully normalized cpg because the SCG-based genome estimate is unavailable.

---

Main outputs

Path	Description
{out_dir}/fastp_summary.csv	Per-sample fastp metrics and Nonpareil coverage
{out_dir}/assembly_qa.tsv	Per-sample assembly QC metrics
{out_dir}/contig_summary.tsv	Per-contig annotations and abundance
{out_dir}/AMR_abundance_summary.csv	Per-sample AMR abundance and risk scores
{out_dir}/mag_summary.tsv	Per-MAG summary table
{out_dir}/data/QAQC/short_reads_output.csv	Short-read AMR gene abundances
{out_dir}/data/QAQC/markers_cpg.csv	pBI143 and crAss001 copies per genome

AMR_abundance_summary.csv contains:

Column	Meaning
sample	Sample ID
AMR_total_cpg	Total AMR abundance across detected genes
pBI143_cpg, crAss001_cpg	Anthropogenic marker abundance
E_exposure	Exposure score from marker abundance
R_mean	Mean resistance-hazard score
M_mean	Mean mobility score
H_mean	Mean host/pathogenicity score
amr_risk_additive_raw, amr_risk_multiplicative_raw	Raw study-level risk scores
amr_risk_additive, amr_risk_multiplicative	Min-max normalized 0-100 scores

---

Abundance and risk scoring

All AMR and contig abundances are reported as copies per genome (cpg):

n_genomes = Σ(alignment_length / gene_length) / 40
cpg       = mean_depth / n_genomes

The 40-gene denominator comes from the bundled single-copy gene reference in workflow/resources/SCGs_40_All.fasta.

Risk scoring combines four components:

• R - resistance hazard from AMRFinderPlus class/subclass
• M - mobility from contig type and nearby MGE evidence
• H - host/pathogenicity from MMseqs2 taxonomy
• E - exposure from pBI143 and crAss001

Both additive and multiplicative sample-level scores are reported.

---

Repository layout

config/                 user configuration and SLURM profiles
docs/                   rule graph and session log
test/                   mock data, mini databases, reproducible test scripts
workflow/
  Snakefile             workflow entry point
  envs/                 per-rule conda environments
  resources/            bundled reference files
  rules/                short-read, contig, MAG, and summary rules
  scripts/              Python/R scripts used by rules
dbs/                    auto-managed databases (gitignored)