The Evoformer is a deep learning module introduced and used by AlphaFold2 that processes Multiple sequence alignments for protein structure prediction. It contains the majority of trainable parameters and is similar to MSA Transformer (probably independently developed) in that it uses axial attention.
Mentions
- The evoformer partially learns protein dynamics (1). Its embeddings can be used without fine-tuning as inputs to a GNN-Transformer hybrid that modeled protein dynamics (albeit imperfectly).
Ref (1)
1.
Zheng S, He J, Liu C, Shi Y, Lu Z, Feng W, et al. Towards Predicting Equilibrium Distributions for Molecular Systems with Deep Learning. 2023; Available from: https://arxiv.org/abs/2306.05445