nucleobench 0.1.2

Optimizers from the nucleobench package.

NucleoBench

This is the initial repo for an upcoming paper, NucleoBench: A Large-Scale Benchmark of Neural Nucleic Acid Design Algorithms.

This repo is covered by the MIT license.

This repo is intended to be used in a few ways:

1. Reproducing the results from our paper.
2. Running the NucleoBench sequence designers on custom problems.
3. Using our new designer, AdaBeam, on a custom problem.

To do these, you can clone this repo, use the Docker image (for the benchmark), or use the PyPi package for our designers.

Results

Installation from PyPi

pip install nucleobench

Then you can use it in python:

from nucleobench import optimizations
from nucleobench import optimizations
opt = optimizations.get_optimization('beam_search_unordered')  # Any optimizer name.

Installation & testing from GitHub

# Clone the repo.
git clone https://github.com/move37-labs/nucleobench.git
cd nucleobench

# Create and activate the conda environment.
conda env create -f environment.yml
conda activate nucleobench

# Run all the unittests.
pytest nucleobench/

You can also run the integration tests, which require an internet connection:

pytest docker_entrypoint_test.py

Nucleic acid design benchmark.

Running NucleoBench

See the folder recipes for examples of how to run the designer locally.

Building a Docker image

To help deploy NucleoBench to the cloud, we've created a docker container. To build it yourself, see the top of Dockerfile for instructions. One way of creating a docker file is:

docker build -t nucleobench -f Dockerfile .