Protein Fold Classification at Scale: Benchmarking and Pretraining

Researchers introduced TEDBench, a large non-redundant benchmark for protein fold classification built from TED and clustered AlphaFold structures.

They also propose MiAE, a masked self-supervised model with high masking and SE(3)-invariant encoding to reconstruct protein backbones.

MiAE scales better than prior approaches and delivers strong performance on TEDBench as well as curated experimental CATH v4.4 data.

The work tackles a major bottleneck in protein structure learning by improving both benchmarking quality and pretraining for better transfer to experimental structures.