LM-GVP: an extensible sequence and structure informed deep learning framework for protein property prediction
Problem
Predicting protein properties such as fluorescence, protease stability, and function from sequence and structure, to speed protein engineering and drug development.
Why it matters
Accurate property prediction could greatly reduce the cost and time of developing bio-therapeutics.
What the AI did
Built LM-GVP, combining a protein language model with a graph neural network to use both 1D sequence and 3D structural information; it outperformed state-of-the-art protein LMs on benchmark prediction tasks.
Results
Outperformed state-of-the-art protein LMs on fluorescence, protease stability, and Gene Ontology function prediction; no numeric metrics stated in abstract.
Impact
Peer-reviewed framework with 85 citations; authors envision generalization to accelerate protein engineering.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Zichen Wang, Steven A. Combs, Ryan M. Brand, Miguel Romero Calvo, Panpan Xu, George Price, Nataliya Golovach, Emmanuel Oluwatobi Salawu et al. |
| Citations | 85 (OpenAlex, at import time) |
| References | 38 |
| Model or method | LM-GVP (protein Language Model + Graph Neural Network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-022-10775-y |
| Open access | gold |
| Keywords | protein language model, graph neural network, property prediction, deep learning, protein engineering |
Importance rating
significant — thousands to millions affected
Abstract states property prediction is invaluable and would greatly accelerate protein engineering and drug development; no scale stated.