fastembed-bio 0.1.2

Fast, light, accurate library for biological sequence embeddings (proteins, DNA, RNA)

fastembed-bio

Fast, lightweight biological sequence embeddings using ONNX. Built on FastEmbed.

Why fastembed-bio?

1. Light: No GPU required. No PyTorch. Just ONNX Runtime. Perfect for serverless and resource-constrained environments.

2. Fast: ONNX Runtime is faster than PyTorch inference. Batch processing and parallelism built-in.

3. Simple: Same interface patterns as FastEmbed. If you've used FastEmbed for text, you already know how to use this.

Installation

pip install fastembed-bio

Quickstart

from fastembed.bio import ProteinEmbedding

sequences = [
    "MKTVRQERLKSIVRILERSKEPVSGAQLAEELSVSRQVIVQDIAYLRSLGYNIVATPRGYVLAGG",
    "GKGDPKKPRGKMSSYAFFVQTSREEHKKKHPDASVNFSEFSKKCSERWKTMSAKEKGKFEDMAK",
]

model = ProteinEmbedding("facebook/esm2_t12_35M_UR50D")
embeddings = list(model.embed(sequences))

# [
#   array([-0.0055, -0.0144,  0.0355, -0.0049, ...], dtype=float32),
#   array([ 0.0114,  0.0020, -0.0247,  0.0060, ...], dtype=float32)
# ]

Supported Models

Protein Embeddings

Model	Parameters	Dimensions	Description
facebook/esm2_t12_35M_UR50D	35M	480	ESM-2 protein language model

from fastembed.bio import ProteinEmbedding

model = ProteinEmbedding("facebook/esm2_t12_35M_UR50D")
embeddings = list(model.embed(["MKTVRQERLKS", "GKGDPKKPRGK"]))

DNA Embeddings (Coming Soon)

DNABert and similar models for DNA sequence embeddings.

RNA Embeddings (Coming Soon)

RNA foundation models for RNA sequence embeddings.

GPU Support

from fastembed.bio import ProteinEmbedding

model = ProteinEmbedding(
    "facebook/esm2_t12_35M_UR50D",
    providers=["CUDAExecutionProvider"]
)

Requires onnxruntime-gpu instead of onnxruntime.

Relationship to FastEmbed

This project is a community-driven fork of FastEmbed focused on biological sequence embeddings. It uses the same core infrastructure (ONNX models, model management, etc.) but is specialized for proteins, DNA, and RNA.

The goal is to make biological embeddings as accessible and efficient as text embeddings.

Contributing

Contributions welcome! Areas of interest:

• Additional ESM-2 model sizes
• DNABert and other DNA models
• RNA foundation models
• Performance optimizations

License

Apache 2.0