Paula Ruiz-Rodriguez¹ and Mireia Coscolla¹
_{1. Institute for Integrative Systems Biology, I²SysBio, University of Valencia-CSIC, Valencia, Spain}

SNPick extracts variable (SNP) sites from whole-genome FASTA alignments. It produces reduced alignments ready for phylogenetic inference with ascertainment bias correction (ASC) in IQ-TREE and RAxML, and optionally generates VCF files.

Why not snp-sites? snp-sites works well for small datasets but struggles with large alignments — it loads everything into memory and scales poorly. SNPick uses a zero-copy memory-mapped architecture that handles thousands of genomes in seconds with minimal RAM.

# Install
conda install -c bioconda snpick

# Extract variable sites
snpick -f alignment.fasta -o snps.fasta

# With VCF output
snpick -f alignment.fasta -o snps.fasta --vcf

# Include gaps as informative
snpick -f alignment.fasta -o snps.fasta -g

Identifies positions with more than one observed nucleotide across all sequences. Constant and ambiguous-only positions are excluded from the output.

Reports constant site counts (fconst) directly, formatted for IQ-TREE's +ASC models:

[snpick] ASC fconst: 744123,1382922,1382180,743556

Use in IQ-TREE:

iqtree2 -s snps.fasta -m GTR+ASC -fconst 744123,1382922,1382180,743556

Optional VCF v4.2 output with per-sample genotypes. Reference allele taken from the first sequence. Ambiguous bases reported as missing (.).

• Ambiguous bases (N, R, Y, etc.): not counted as alleles — positions are only variable if they have ≥2 standard bases (A, C, G, T)
• Gaps (-): ignored by default, included as a 5th character with -g

Automatic multi-threaded scanning via Rayon when the dataset is large enough. Falls back to single-threaded for small inputs to avoid overhead.

conda install -c bioconda snpick
# or
mamba install -c bioconda snpick

git clone https://github.com/PathoGenOmics-Lab/snpick.git
cd snpick
cargo build --release
# Binary at target/release/snpick

wget https://github.com/PathoGenOmics-Lab/snpick/releases/latest/download/snpick
chmod +x snpick

snpick [OPTIONS] --fasta <FASTA> --output <OUTPUT>

Input (alignment.fasta):

>sequence1
ATGCTAGCTAGCTAGCTA
>sequence2
ATGCTAGCTGGCTAGCTA
>sequence3
ATGCTAGCTAGCTAGCTA

Command:

snpick -f alignment.fasta -o snps.fasta

Output (snps.fasta):

>sequence1
A
>sequence2
G
>sequence3
A

stderr:

[snpick] Mapped 63 bytes. 3 sequences × 18 positions.
[snpick] 1 variable, 17 constant (A:4 C:4 G:4 T:5), 0 ambiguous-only, 18 total.
[snpick] ASC fconst: 4,4,4,5
[snpick] Done in 0.00s. 1 vars from 3 seqs × 18 pos.

Simulated M. tuberculosis-like genomes (4.4 Mbp, ~65% GC, 3.6% variable sites).

SNPick maintains O(L) memory regardless of sequence count, while snp-sites requires O(N×L).

Input FASTA ──mmap──▶ Index records ──▶ Pass 1: bitmask scan ──▶ Analyze
                           │                    (parallel)           │
                           │                                         ▼
                           └──────────▶ Pass 2: extract sites ──▶ FASTA + VCF
                                         (sparse random access)

• Single memory-mapped file shared across both passes — zero copies
• Pass 1: OR-based bitmask over all sequences (parallel with Rayon)
• Pass 2: only reads variable positions (sparse access via mmap)
• Lookup tables: 256-byte arrays for O(1) nucleotide classification and case conversion

If you use SNPick in your research, please cite:

@software{snpick,
  author    = {Ruiz-Rodriguez, Paula and Coscolla, Mireia},
  title     = {SNPick: Fast extraction of variable sites from FASTA alignments},
  url       = {https://github.com/PathoGenOmics-Lab/snpick},
  doi       = {10.5281/zenodo.14191809},
  license   = {GPL-3.0}
}

Paula Ruiz-Rodriguez 💻 🔬 🤔 🔣 🎨 🔧

Mireia Coscolla 🔍 🤔 🧑‍🏫 🔬 📓

This project follows the all-contributors specification (emoji key).