meta2vec is a Python package for metabolite embedding, which allows for the representation of metabolites in a vector space.

meta2vec package contains three modules:

• distance: provides functions for calculating the similarity distance between two metabolites using their embeddings.
• utils: provides helper functions for working with HMDB (Human Metabolome Database) dataset.
• visualize: provides functions for visualizing the embeddings using UMAP.

An example.ipynb is provides.

meta2vec package can be installed via pip:

pip install meta2vec

Here is a brief overview of how to use meta2vec package:

from meta2vec import from_pretrained

# Load the TransE pre-trained embeddings
hmdb_embeddings = from_pretrained(model_type="TransE", embedding_dir="embedding")

The above code loads the pre-trained HMDB embeddings from the embedding directory. If the embeddings are not already present in the directory, the function will download them from the GitHub repository.

from meta2vec import hmdb_id_cosine_distance, hmdb_id_euclidean_distance

# Calculate the cosine distance between two HMDB IDs
cosine_distance = hmdb_id_cosine_distance(hmdb_embeddings, "HMDB0000001", "HMDB0000002")

# Calculate the Euclidean distance between two HMDB IDs
euclidean_distance = hmdb_id_euclidean_distance(hmdb_embeddings, "HMDB0000001", "HMDB0000002")

from meta2vec import most_similar

# Find the most similar HMDB IDs to a given HMDB ID
similar_compounds = most_similar(hmdb_embeddings, "HMDB0000001")

from meta2vec import visualize_umap

# Visualize the embeddings using UMAP
visualize_umap(hmdb_embeddings)

If you would like to contribute to this project, please contact [email protected]