How Machine Learning tools at Google made AlphaFold possible
summary
AlphaFold uses deep learning with attention mechanisms to predict protein 3D structure from sequence and is judged at CASP as a general solution to the folding problem. The model is trained on 128 TPU v3 cores and reaches production quality after trained in a couple of weeks, was developed on TensorFlow 1 then migrated and open‑sourced JAX code and weights, runs in seconds to minutes on TPU/GPU and powers a public database of all human proteins plus 20 organisms.
AI-written summary · about
chapters (6)
- 0:00Intro
- 0:20Proteins
- 0:53How does AlphaFold work
- 1:13How does AlphaFold make predictions
- 1:31Research
- 1:52Training
related talks
36mAlgorithmic Alchemy: Transforming Drug Discovery with Open AI
01h 13mDeep Learning in Structural Biology and Protein Design: How, Where, and Why
29mProtein folding and what it means for drug discovery
53mHow machine learning is influencing protein engineering
14mAI Is Built to Be Applied
06mTools vs Products: Lessons learned from a niche industry
14mAI for Science
12mKeynote: What's New in AI