DISCLAIMER: this space is designed for research purposes only. Its accuracy has not been tested and it does not compare to the one of AlphaFold3. However, the core concept of protein folding prediction and the intuitive user interface make it a good alternative to Google Deepmind's tool
This is the code for the HuggingFace space: as-cle-bert/proteinviz, make sure to check it out!
proteinviz lets you predict the 3D structure of a protein starting from its amino-acidic sequence, as you can see here:
It does this by exploiting facebook/esmfold_v1 protein folding model by Facebook-Meta, available on HuggingFace Hub.
The model predicts the positions of the amino-acids and its predictions get turned into a PDB file (Protein DataBase format). This file contains lots of stereochemical and positional metadata related to amino-acids and their atoms, instructing visualization servers on how the protein should be represented.
The protein is then visualized at an amino-acidic level thanks to gradio_molecule3d component by Simon Dürr. The tertiary structure 3D representation of the molecule is instead written to an HTML object that gets displayed on your browser, directly. Both these things are accomplished thanks to 3Dmol.js (more at Citations)
Everything is easily rendered through a Gradio interface.
⚠️ WARNING: You should reproduce this Gradio app if and only if you have a good GPU, otherwise it might crash
You first need to clone this repo:
git clone https://github.com/AstraBert/proteinviz.gitAfter that, you navigate to the cloned repository and install the needed dependencies:
python3 -m pip install -r requirements.txtNow that you have everything set, you can launch the application:
python3 app.pyYou should see the application running on localhost:7860 or 0.0.0.0:7860 after a while.
The hereby presented software is open-source and distributed under MIT license.
As stated before, the project was developed for research purposes and must be used only in this sense.
- Rego N, Koes D. 3Dmol.js: molecular visualization with WebGL. Bioinformatics. 2015;31(8):1322-1324. doi:10.1093/bioinformatics/btu829