SweepGA

Fast all-vs-all genome alignment with plane-sweep filtering and scaffold-aware chaining. Wraps FastGA (default) and wfmash, then applies a plane sweep to drop near-duplicate mappings and (optionally) chains nearby alignments into syntenic scaffolds.

What it does

• Align — one or more FASTA files → PAF or .1aln. Uses FastGA or wfmash; PanSN-aware; scales k-mer frequency to the cohort size automatically.
• Filter — take any aligner's PAF/1aln and apply the same plane-sweep
  • scaffold pipeline.
• Scale — built-in batching (--batch-bytes, --max-disk) partitions large cohorts into memory/disk-bounded batches.
• Sparsify — optionally pre-select which pairs to align (random, tree, giant-component, wfmash-density).

Ships two binaries: sweepga (the aligner/filter) and alnstats (a small PAF/1aln statistics tool).

Install

# From crates.io
cargo install sweepga

# From source
git clone https://github.com/pangenome/sweepga.git
cd sweepga
cargo install --force --path .

Requires Rust ≥ 1.70.

On mixed-glibc systems (e.g. Debian + Guix) a plain cargo build may fail with __pthread_barrier_wait@GLIBC_PRIVATE link errors. Use ./scripts/build-clean.sh --install — see docs/BUILD-NOTES.md.

Quick start

# Self-alignment of one FASTA (FastGA, default)
sweepga genome.fa.gz --output-file aln.paf

# Use wfmash instead
sweepga --wfmash genome.fa.gz --output-file aln.paf

# Pairwise alignment of two FASTAs
sweepga target.fa query.fa --output-file aln.paf

# Filter an existing PAF (any aligner)
sweepga alignments.paf --output-file filtered.paf

# Convert .1aln → PAF
sweepga alignments.1aln --paf > out.paf

# AGC archive input (extract samples, align, filter)
sweepga --agc-samples HG002,HG005 genomes.agc --output-file aln.paf

sweepga auto-detects input type (FASTA vs. PAF/1aln) and picks a reasonable pipeline. Run sweepga --help for the full flag list.

Defaults

The defaults are tuned for pangenome alignment (matches impg's historical settings):

Flag	Default	Meaning
--aligner	fastga	Aligner backend. Use --wfmash or --aligner wfmash to switch.
--num-mappings	many:many	Pre-scaffold plane-sweep: keep all mappings per query/target.
--scaffold-jump	50k	Scaffolding is enabled by default; chains mappings within a 50kb gap. Pass --scaffold-jump 0 to disable.
--scaffold-mass	10k	Drop scaffold chains shorter than 10kb.
--scaffold-filter	many:many	Keep all non-overlapping scaffolds. Use 1:1 for the aggressive (best-per-chromosome-pair) filter.
--overlap	0.95	Drop mappings whose overlap with a better-scoring one exceeds 95%.
--scoring	log-length-ani	Plane-sweep scoring function.
--threads	8	Parallelism.

Adaptive behavior: when the input average sequence length is known (FASTA with .fai), --scaffold-mass and --scaffold-jump are clamped to the sequence-length scale so short inputs (e.g. pangenome-window excerpts) don't get filtered empty. Pass --no-adaptive-scaffolds to turn clamping off.

Scaffolding pipeline

When --scaffold-jump > 0 (the default), sweepga runs:

1. Pre-scaffold plane sweep per (query, target) chromosome pair using --num-mappings / --scoring / --overlap.
2. Scaffold creation — union-find merges mappings within --scaffold-jump on both axes; chains shorter than --scaffold-mass are dropped.
3. Scaffold plane sweep using --scaffold-filter (default many:many); 1:1 keeps the single best scaffold per chromosome-pair.
4. Rescue (--scaffold-dist > 0) — recover mappings within the Euclidean distance of a kept scaffold anchor. Off by default.

Pass --no-filter to skip everything and just emit the raw aligner output (or the raw input PAF).

Sparsification

--sparsify picks which pairs actually get aligned (for FASTA input) or how densely wfmash reports mappings:

Value	Behavior
none (default)	Align all pairs, report all mappings
auto	Heuristic pick based on cohort size
<f> or random:<f>	Random <f> fraction of pairs
giant:<p> / connectivity:<p>	Giant-component connectivity guarantee
tree:<near>[:<far>[:<random>]]	Tree-sampled neighbor pairs
wfmash:auto / wfmash:<f>	wfmash mapping density (-x flag); auto = ln(n)/n*10

--sparsify-pairs is a separate knob that only drives pre-alignment pair selection (same grammar, no wfmash variant).

PanSN

Sequence names like SAMPLE#HAPLOTYPE#CHR are treated pair-wise per-chromosome within each genome pair. Prefix delimiter is #.

Output

• .paf — standard PAF text.
• .1aln — compact binary ONE format (50–70% smaller, preserves edit distances). Pass --paf to convert to PAF on read.

Output path is chosen by --output-file (extension auto-detected) or the explicit --paf / --1aln flags.

Alnstats

alnstats alignments.paf              # summary
alnstats raw.paf filtered.paf        # before/after
alnstats alignments.paf -d           # per-genome-pair breakdown

License

MIT.