https://github.com/google-research/slip
SLIP is a sandbox environment for engineering protein sequences with synthetic fitness functions.
https://github.com/google-research/slip
computational-biology machine-learning protein-design
Last synced: 6 months ago
JSON representation
SLIP is a sandbox environment for engineering protein sequences with synthetic fitness functions.
- Host: GitHub
- URL: https://github.com/google-research/slip
- Owner: google-research
- Archived: true
- Created: 2021-11-24T19:36:36.000Z (almost 4 years ago)
- Default Branch: main
- Last Pushed: 2024-01-17T07:29:23.000Z (over 1 year ago)
- Last Synced: 2025-04-03T01:01:49.788Z (6 months ago)
- Topics: computational-biology, machine-learning, protein-design
- Language: Python
- Homepage:
- Size: 132 KB
- Stars: 19
- Watchers: 3
- Forks: 9
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
Awesome Lists containing this project
README
This is not an officially supported Google product.
# SLIP - Synthetic Landscape Inference for Proteins
SLIP is a sandbox environment for engineering protein sequences with
synthetic fitness functions. See our [preprint](https://www.biorxiv.org/content/10.1101/2022.10.28.514293v1)
## Installation instructions
Tested on python >= 3.7
We recommend installing into a [virtual environment](https://docs.python.org/3/library/venv.html) to isolate dependencies.
```
python3 -m venv env
source env/bin/activate
```
To install:
```
pip3 install -q -r requirements.txt
```
To run the unit tests:
```
bash -c 'for f in *_test.py; do python3 $f || exit 1; done'
```
## Example landscape usage
See this [colab](https://colab.research.google.com/drive/1BkR2KvvjgzUTJg5VO3BsuTPSDjQisnbJ) for an example of using a landscape.
## Constructing a new landscape
All landscapes were constructed using [Mogwai](https://github.com/songlab-cal/mogwai). See that repo's [example](https://github.com/songlab-cal/mogwai/blob/main/examples/gremlin_train.ipynb), which shows how to train a new Potts model and how to (optionally) examine contact accuracy after training. All that is required is an alignment in .a3m format, true contacts are not required (e.g. as in this [colab](https://github.com/songlab-cal/slc22a5/blob/main/slc22a5_train_potts.ipynb)).