Machine Learning-based Design of Proteins and Small Molecules
summary
Frames protein and small‑molecule design as a practical model-based optimization problem that replaces slow assays with a deep‑net stochastic oracle, then searches by optimizing a generative search model (VAE/HMM style) via weighted maximum‑likelihood and annealed rare‑event estimation (cross‑entropy method). It adds a learned or curated density p(x) — conditioning on trustworthy p(x) — to avoid oracle black‑holes, uses GP‑based simulation for validation, and reports wet‑lab collaborations but no finalized lab successes yet.
AI-written summary · about
chapters (22)
- 0:00Intro
- 1:28Why do we want to do this
- 2:34Proteins
- 5:45Directed Evolution
- 12:23How to think about this
- 13:20First approach
- 13:56Modelbased optimization
- 21:55Distribution estimation
- 25:10Challenges
- 27:08Black Holes
- 27:50Panda
- 28:02Gibbon
- 28:14Banana
- 28:31Image Generation
- 38:02Simulations
- 38:49Model Based Optimization
- 39:07Collaborations
- 39:46Extrapolation
- 40:49Domain Adaptation
- 41:17Uncertainty Calibration
- 41:48Deep Neural Networks
- 42:41Expectationmaximization
related talks
32mMachine Learning-Based Design Of Proteins
53mHow machine learning is influencing protein engineering
01h 13mDeep Learning in Structural Biology and Protein Design: How, Where, and Why
31mML methods to accelerate the drug discovery cycle
43mMachine Learning in Molecular Biology
16mBeyond Patterns: The Case for Physics-Grounded AI in Drug Discovery
30mMachine Learning: A New Approach to Drug Discovery
11mImproving Drug Discovery with Machine Learning...