Next Generation Protein Simulation
OpenProtein is a new machine learning framework for modeling tertiary protein structure. Built on top of PyTorch, NGL Viewer and ProteinNet, it offers automatic differentiation, visualisation, built-in data sets and much more.
We introduce a new machine learning framework, OpenProtein, for prediction of tertiary protein structure. The framework is based upon PyTorch and utilizes NGL Viewer for live visualization of the protein as it folds. OpenProtein parses the ProteinNet data set (derived from previous CASP competitions) directly using a memory efficient data loader and supports a wide range of features including built-in embedding, automatic differentiation and performance plotting.
We know how important quick iteration speed is in research. OpenProtein gives you feedback on your model changes in seconds.
All OpenProtein models are GPU compatible by default and the framework's custom PyTorch data loaders are optimized for memory efficiency.
Simply focus on designing your model (in many cases less than 50 lines of code!) and the framework will handle the rest for you, including training and validation.
Frequently Asked Questions
Is this yet another structure prediction tool?
Quite the opposite in fact - OpenProtein is a framework for creating, training and evaluating models for protein structure prediction. It can be used to improve existing models too!
What dependencies do I need to run OpenProtein?
Open Protein runs in any Python 3 environment with PyTorch installed. A GPU is recommended, but not required.
How do I contribute to OpenProtein?
Simply create a pull request in the project's Github repository at https://github.com/OpenProtein/openprotein