Bio Embeddings

The project includes:

• A pipeline that allows to embed a FASTA file choosing from various embedders (see below), and then project and visualize the embeddings on 3D plots.
• A web server that takes in sequences, embeds them and returns the embeddings OR visualizes the embedding spaces on interactive plots online.
• General purpose library to embed protein sequences in any python app.

We presented the bio_embeddings pipeline as a talk at ISMB 2020. You can find it on YouTube, and a copy of the poster will soon be available on F1000.

Important information

• The albert model weights are not publicly available yet. You can request early access by opening an issue.
• Please help us out by opening issues and submitting PRs as you see fit, this repository is actively being developed.

Install guides

You can install the package via pip like so:

pip install bio-embeddings

Or directly from the source (e.g. to have the latest features):

pip install -U git+https://github.com/sacdallago/bio_embeddings.git

Examples

We highly recommend you to check out the examples folder for pipeline examples, and the notebooks folder for post-processing pipeline runs and general purpose use of the embedders.

After having installed the package, you can:

1. Use the pipeline like:

   bio_embeddings config.yml
   

   A blueprint of the configuration file, and an example setup can be found in the examples directory of this repository.

2. Use the general purpose embedder objects via python, e.g.:

   from bio_embeddings import SeqVecEmbedder
   
   embedder = SeqVecEmbedder()
   
   embedding = embedder.embed("SEQVENCE")
   

   More examples can be found in the notebooks folder of this repository.

Development status

1. Pipeline stages

   • embed:
     • SeqVec v1/v2 (https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-019-3220-8)
     • Fastext
     • Glove
     • Word2Vec
     • UniRep (https://www.nature.com/articles/s41592-019-0598-1?sfns=mo)
     • Albert (unpublished)
   • project:
     • t-SNE
     • UMAP

2. Web server (unpublished):

   • SeqVec
   • Albert (unpublished)

3. General purpose objects:

   • SeqVec
   • Fastext
   • Glove
   • Word2Vec
   • UniRep
   • Albert (unpublished)

Building a Distribution

Building the packages best happens using invoke. If you manganage your dependecies with poetry this should be already installed. Simply use poetry run invoke clean build to update your requirements according to your current status and to generate the dist files

Additional dependencies and steps to run the webserver

If you want to run the webserver locally, you need to have some python backend deployment experience. You'll need a couple of dependencies if you want to run the webserver locally: pip install dash celery pymongo flask-restx pyyaml.

Additionally, you will need to have two instances of the app run (the backend and at least one celery worker), and both instances must be granted access to a MongoDB and a RabbitMQ or Redis store for celery.

Contributors

• Christian Dallago (lead)
• Konstantin Schütze
• Tobias Olenyi
• Michael Heinzinger