Deep learning approaches to predict protein 3D structure from sequence alone, achieving experimental accuracy. Explores attention mechanisms, multiple sequence alignments, and iterative refinement for structural biology applications.
Core Beliefs
Protein structure is determined by amino acid sequence
Evolutionary information from homologs constrains possible structures
Attention mechanisms can learn spatial relationships in 3D structures
Methods
Transformer architectures with self-attention over residue pairs
Multiple sequence alignments and co-evolution analysis