msasim 25.11.11

A fast MSA simulator

Sailfish

Sailfish is a high-performance multiple sequence alignment (MSA) simulator written in C++ with a Python API. It enables rapid generation of large-scale simulated datasets with support for indels, substitutions, and realistic evolutionary models.

Features

• High-performance C++ engine with ergonomic Python interface
• Support for both DNA and protein sequence evolution
• Flexible indel modeling with multiple length distributions (Zipf, Geometric, Poisson, Custom)
• 26+ substitution models including JTT, WAG, LG, HKY, GTR, and more
• Gamma rate heterogeneity and invariant sites
• Per-branch parameter specification for heterogeneous models
• Low-memory mode for large-scale simulations (1M+ sequences)
• Reproducible simulations with explicit seed control

Installation

pip install msasim

Requirements: Python >= 3.6

Quick Start

Basic Example with Indels and Substitutions

from msasim import sailfish as sim
from msasim.sailfish import MODEL_CODES, ZipfDistribution

# Configure simulation protocol
sim_protocol = sim.SimProtocol(
    tree="(A:0.5,B:0.5);",
    root_seq_size=100,
    deletion_rate=0.01,
    insertion_rate=0.01,
    deletion_dist=ZipfDistribution(1.7, 50),
    insertion_dist=ZipfDistribution(1.7, 50),
    seed=42
)

# Create simulator
simulation = sim.Simulator(sim_protocol, simulation_type=sim.SIMULATION_TYPE.PROTEIN)

# Configure substitution model with gamma rate heterogeneity
simulation.set_replacement_model(
    model=MODEL_CODES.WAG,
    gamma_parameters_alpha=1.0,
    gamma_parameters_categories=4
)

# Run simulation
msa = simulation()

# Output results
msa.write_msa("output.fasta")
msa.print_msa()

Substitutions-Only Simulation

from msasim import sailfish as sim
from msasim.sailfish import MODEL_CODES

# No indels configured
protocol = sim.SimProtocol(
    tree="path/to/tree.nwk",
    root_seq_size=500,
    seed=42
)

simulator = sim.Simulator(protocol, simulation_type=sim.SIMULATION_TYPE.PROTEIN)
simulator.set_replacement_model(model=MODEL_CODES.LG)

msa = simulator()
msa.write_msa("alignment.fasta")

Batch Simulations

from msasim import sailfish as sim
from msasim.sailfish import MODEL_CODES

# Initialize once with seed
protocol = sim.SimProtocol(tree="tree.nwk", root_seq_size=500, seed=42)
simulator = sim.Simulator(protocol, simulation_type=sim.SIMULATION_TYPE.PROTEIN)
simulator.set_replacement_model(model=MODEL_CODES.JTT)

# Generate multiple replicates
# Internal RNG advances automatically for reproducibility
for i in range(100):
    msa = simulator()
    msa.write_msa(f"replicate_{i:04d}.fasta")

Low-Memory Mode for Large Simulations

import pathlib
from msasim import sailfish as sim
from msasim.sailfish import MODEL_CODES

protocol = sim.SimProtocol(tree="large_tree.nwk", root_seq_size=10000, seed=42)
simulator = sim.Simulator(protocol, simulation_type=sim.SIMULATION_TYPE.DNA)
simulator.set_replacement_model(model=MODEL_CODES.NUCJC)

# Write directly to disk without holding MSA in memory
simulator.simulate_low_memory(pathlib.Path("large_alignment.fasta"))

Documentation

For complete API documentation, including all available models, distributions, and advanced features, see API_REFERENCE.md.

Core Concepts

Simulation Types

• SIMULATION_TYPE.NOSUBS: Indels only, no substitutions
• SIMULATION_TYPE.DNA: DNA sequences with nucleotide models
• SIMULATION_TYPE.PROTEIN: Protein sequences with amino acid models

Available Models

Nucleotide: JC, HKY, GTR, Tamura92 Protein: WAG, LG, JTT (JONES), Dayhoff, MTREV24, CPREV45, HIV models, and more

See API_REFERENCE.md#substitution-models for the complete list.

Indel Length Distributions

• ZipfDistribution: Power-law distribution (typical for biological data)
• GeometricDistribution: Exponentially decreasing
• PoissonDistribution: Poisson-based
• CustomDistribution: User-defined probability vector

Common Use Cases

1. Phylogenetic Method Validation

Generate known-truth alignments with specified evolutionary parameters to test inference methods.

2. Benchmarking Alignment Tools

Create challenging datasets with varying indel rates and substitution patterns.

3. Statistical Power Analysis

Simulate datasets under different evolutionary scenarios to assess method sensitivity.

4. Model Comparison

Generate alignments under different substitution models for model selection studies.

Performance Notes

Typical simulation times on modern hardware:

• 10K sequences × 30K sites: ~10 seconds
• 100K sequences × 30K sites: ~2 minutes
• 1M sequences × 30K sites: ~20 minutes

For simulations with >100K sequences or memory constraints, use simulate_low_memory().

Memory usage estimate: (num_sequences × alignment_length) / 300,000 MB

Project Goals

• Ease of use: Simple, intuitive Python API
• Speed: High-performance C++ implementation
• Modularity: Flexible configuration of all evolutionary parameters

Contributing

Bug reports and feature requests are welcome via GitHub issues.

Citation

If you use Sailfish in your research, please cite:

[Citation information to be added]

License

[License information to be added]