How machine learning is influencing protein engineering
summary
Field reports that AlphaFold and inverse folding models enable structure-to-sequence pipelines but fail for enzymes and dynamic conformations; the work shows practical pipelines combining pan-protein language models, conditional generative models, and iterative model-based optimization that uses an EM-like update with a soft trust region and importance reweighting to control domain shift. Zero-shot ranking from evolutionary MSAs guides initial libraries, synthesis costs run roughly $5–$10 per short sequence so teams design distributions not points, and Bayesian priors plus diffusion guidance tweaks improve extrapolative reliability.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:11Protein engineering basics
- 7:08Machine learning for proteins
- 10:16AlphaFold in protein design
- 13:23Prediction vs design tasks
- 16:36ML trends in proteins
- 21:34Protein function prediction
- 24:47AlphaFold gaps and future
- 28:42Model-based optimization project
- 37:48Library design and synthesis
- 41:56Iterative optimization and domain shift
- 49:20Other projects and closing
related talks
01h 13mDeep Learning in Structural Biology and Protein Design: How, Where, and Why
29mProtein folding and what it means for drug discovery
04mHow Machine Learning tools at Google made AlphaFold possible
46mMachine Learning-based Design of Proteins and Small Molecules
36mAlgorithmic Alchemy: Transforming Drug Discovery with Open AI
32mMachine Learning-Based Design Of Proteins
06mTools vs Products: Lessons learned from a niche industry
16mBeyond Patterns: The Case for Physics-Grounded AI in Drug Discovery