bio_ml_models

Collection of prediction models for biological tasks.

About

bio_ml_models provides an easy-to-use interface to use current, well-tested prediction models for a range of biological tasks. bio_ml_models is designed to be easy to use and understand, well-documented and tested. Sharing your bio-research model was never that easy before!

Available Tasks

Proteins

• adeno_associated_virus_fitness: Predict a fitness score for each AAV capsid protein mutation. (1 model(s) available)
• conservation: Predict a conservation score for each residue in a protein sequence. (1 model(s) available)
• gb1_binding: Predict a binding score for each variant of the GB1 protein. (1 model(s) available)
• meltome: Predict the meltome temperature value of a protein. (1 model(s) available)
• multi_ligand_binding_site: Predict binding sites of ligands for protein residues. (1 model(s) available)
• secondary_structure: Predict the secondary structure for each residue in a protein sequence. (1 model(s) available)
• single_amino_acid_variant_effect: Predict if a single amino acid variant of a protein sequence has an effect. (1 model(s) available)
• sub_cellular_location: Predict the subcellular location for a protein. (1 model(s) available)

Installation

Make sure that you have python >= 3.9 and poetry installed.

poetry install

Usage

The general entrypoint to use the package is to instantiate the collection. *Note that on the first use, this will download the model data (~250 MB) to your user .cache directory:

from bio_ml_models import BioModelCollection

# Load collection
collection = BioModelCollection()  # Automatically downloads model data 

The default download URL is https://biocentral.cloud/downloads/bio_ml_models/model_data_{version}.zip. If you want to use another source, you can configure that when creating the collection:

from bio_ml_models import BioModelCollection

# Load collection and download data from custom domain
collection = BioModelCollection(custom_download_url="https://example.com/your-archive.zip")  

Here's a complete example how to use bio_ml_models for predictions, using the aav_flip model:

from bio_ml_models import BioModelCollection

from biotrainer.embedders.one_hot_encoding_embedder import OneHotEncodingEmbedder

# Load collection
collection = BioModelCollection()  # Automatically downloads model data

# Attribute-style access
aav_flip = collection.proteins.adeno_associated_virus_fitness.aav_flip

# Dictionary-style access
aav_flip = collection["proteins"]["adeno_associated_virus_fitness"]["aav_flip"]

# Example sequences, shortened from the training set
example_sequences = ["SEPRPIGTRYLTRNL", "EPRPIGTRYLTRNL"]

# Embed
one_hot_encoding_embedder = OneHotEncodingEmbedder()
embeddings = {f"Sequence{i}": one_hot_encoding_embedder.reduce_per_protein(embd) for i, embd in
              enumerate(one_hot_encoding_embedder.embed_many(sequences=example_sequences))}

# Predict and print result
print(aav_flip.predict(input_data=embeddings))

Contributing

We happily welcome contributions!

• Contributing to the code, examples or documentation is always possible via creating an issue or pull request.
• If you want to add a new model or update an existing one, please get in touch.

Citation

A publication about bio_ml_models is under construction. Until then, please cite this repository if you are using bio_ml_models for your work:

@Online{bio_ml_models,
  accessed = {2024-10-01},
  author   = {bio_ml_models contributors},
  title    = {bio_ml_models - Collection of prediction models for biological tasks},
  url      = {https://github.com/biocentral/bio_ml_models},
}