Rosetta decoy dataset #8
Description
Referenced in State-of-the-Art Estimation of Protein Model Accuracy using AlphaFold, the Rosetta decoy dataset is a useful benchmark for single sequence structure predictors, since the paper showed that co-evolutionary information may not be needed to learn a protein structure potential function. Better sequence representations (such as those learned by AminoBERT) should in principle lead to improved decoy ranking.
The dataset is available at this link, and potentially useful code related to it can be found here.