🔭 mininear

A NumPy port of the NEAR code for embedding protein sequences.

🗺️ Overview

NEAR (neural embeddings for amino acid relationships) is a method developed by Daniel Olson et al.[1] to generate meaningfull residue-level embeddings for proteins, which can then be used for nearest-neighbor search in high-dimensional space.

mininear is a pure-Python package to encode protein sequences into NEAR embeddings, using the trained weights from the NEAR ResNet, and a portable re-implementation of ResNet using NumPy (and code derived from the numpy-ml project).

This library only depends on NumPy and is available for all modern Python versions (3.7+).

🔧 Installing

Install the mininear package from GitHub, until a PyPI release is made available:

$ pip install git+https://github.com/althonos/mininear

💡 Example

mininear provides a single Encoder class, which can be used to encode a sequence using the NEAR ResNet:

import mininear

encoder = mininear.Encoder()
embedding = encoder.encode_sequence("MLELLPTAVEGVSQAQITGRPEWIWLALGTALMGLGTL...")

The embedding for that sequence is returned as a NumPy array. Currently there is no batching support, but this is planned if that is a desirable feature.

💭 Feedback

⚠️ Issue Tracker

Found a bug? Have an enhancement request? Head over to the GitHub issue tracker if you need to report or ask something. If you are filing in on a bug, please include as much information as you can about the issue, and try to recreate the same bug in a simple, easily reproducible situation.

🏗️ Contributing

Contributions are more than welcome! See CONTRIBUTING.md for more details.

📋 Changelog

This project adheres to Semantic Versioning and provides a changelog in the Keep a Changelog format.

⚖️ License

This library is provided under the GNU General Public License v3.0. It includes some code adapted from the numpy-ml which is also released under the GPL 3.0. The NEAR code and weights are released under the BSD-3-Clause License and are redistributed and adapted here under those terms.

This project is in no way not affiliated, sponsored, or otherwise endorsed by the original NEAR authors. It was developed by Martin Larralde during his PhD project at the Leiden University Medical Center in the Zeller team.

📚 References

• [1] Daniel Olson, Thomas Colligan, Daphne Demekas, Jack W Roddy, Ken Youens-Clark, Travis J Wheeler, NEAR: neural embeddings for amino acid relationships, Bioinformatics, Volume 41, Issue Supplement_1, July 2025, Pages i449–i457, doi:10.1093/bioinformatics/btaf198.