Protein Fold Classification at Scale: Benchmarking and Pretraining
Researchers have developed TEDBench, a new large-scale benchmark for protein fold classification, designed to overcome limitations in existing datasets and models. To address performance issues with current methods, they introduced Masked Invariant Autoencoders (MiAE), a self-supervised learning framework. MiAE utilizes a high masking ratio and an SE(3)-invariant encoder to effectively learn protein structure representations, outperforming supervised methods on the new benchmark. AI
IMPACT Introduces a new benchmark and a self-supervised framework that could improve protein structure analysis and accelerate biological discovery.