ProteinMPNN

#10 in Wetenschop- & Bio-KI

Baker Lab · 2× · tolest 02. Okt. 2026

47

Momentum

ProteinMPNN is an Open Source deep learning tool for "inverse protein design": it calculates amino acid sequences suited to a given protein backbone structure that fold into this structure. It was developed by the Baker Lab (Institute for Protein Design, University of Washington) and published in Science in 2022. The model is based on a Message-Passing Neural Network (Encoder-Decoder architecture) and achieved 52.4% accuracy in native sequence recovery compared to 32.9% with Rosetta. The code is freely available on GitHub (MIT license) and is used via Colab notebooks and a Hugging Face web interface, among other platforms.

Belege (2)