AlphaFold
Predicts the three-dimensional structure of proteins from their amino-acid sequence
WHAT IT IS
AlphaFold is Google DeepMind’s protein-structure prediction system; the second version (AlphaFold2) was announced at the CASP14 assessment in November 2020 with accuracy close to experimental methods. It takes a protein’s amino-acid sequence and a multiple sequence alignment as input, uses an attention-based Evoformer module to model relationships between residues, and outputs three-dimensional coordinates. DeepMind later open-sourced the code and model weights and, with EMBL-EBI, built the AlphaFold Protein Structure Database, which holds hundreds of millions of predicted structures.
Why it matters
It pushed protein-structure prediction to near-experimental accuracy, released the code and weights, and built a large structure database, becoming one of the most cited cases of AI applied to scientific research.
Key specs
- Open source
- Yes (code and model weights released)
- Database scale
- AlphaFold DB holds hundreds of millions of predicted structures
- Input
- Amino-acid sequence and multiple sequence alignment
- Released
- 2020-11 (CASP14)
Related concepts
Attention Mechanism
Every position can look directly at every other position and dynamically weight how much attention to pay
Training & Inference Infrastructure
Memory decides how large a model you can train, communication how long it takes — raw compute is rarely the bottleneck