Protein Structure Prediction

AI speeds up protein structure prediction models and makes large biomolecular systems easier to model.

Business impact

• Open-source models optimized — Anthropic optimized more than 30 open-source biomolecular models.
• Speedup — Anthropic reports the optimized models ran roughly 4x faster on average.
• Protein structure predictions in the AlphaFold database — DeepMind’s AlphaFold Protein Structure Database provides over 200 million protein structure predictions.

Data requirements

• Amino-acid sequences (text) — AlphaFold predicts protein structures from amino-acid sequences.
• Biomolecular model code (text) — Claude optimizes the code for open-source biological models.

AI methods and techniques

• Protein structure prediction — Models infer three-dimensional protein structure from amino-acid sequences.
• Model optimization — Claude is used to improve inference speed and memory use in open-source biological models.
• Kernel development — FlashPairformer provides custom kernels to speed up triangle attention and triangle multiplication.

AI models and model families

Claude, AlphaFold

Industries

Real-world evidence

2 documented case studies on record.

Companies using this: Anthropic, Google DeepMind.