fasta-embed

Embed your DNA sequences into vector space

Generate dense embeddings for DNA sequences using pluggable model backends
(DNABERT-S, Nucleotide Transformer v2 & 3, or your own).

Installation

pip install -r requirements.txt

Quick start

# Using CLI flags
python -m fasta_embed --embedder ntv3 --input sequences.fasta --output embeddings.npy

# Using a config file
cp config.example.yaml config.yaml   # edit as needed
python -m fasta_embed --config config.yaml

# Mix both (CLI flags override the config file)
python -m fasta_embed --config config.yaml --region V3V4 --device cuda:0

Available embedders

List them at any time:

python -m fasta_embed --list-embedders

Configuration

All options can be set via YAML config file, CLI flags, or both (CLI wins).
See config.example.yaml for the full reference.

Key options:

Adding a custom embedder

1. Create a new file in fasta_embed/embedders/, e.g. my_model.py.
2. Subclass Embedder and decorate with @register:

from fasta_embed.embedders import register
from fasta_embed.embedders.base import Embedder

@register("my-model")
class MyModelEmbedder(Embedder):
    def __init__(self, model_id: str = "default/id", device: str | None = None):
        ...

    def load(self) -> None:
        ...

    def embed(self, sequence: str) -> np.ndarray:
        ...

1. Import the module in fasta_embed/embedders/__init__.py:

from . import my_model  # noqa

1. Use it: python -m fasta_embed --embedder my-model --input seqs.fasta