AI for Science
How AI is changing biology and chemistry: protein folding, AlphaFold, protein language models, protein design, molecules in Python, and AI weather models.
Before this: Deep Learning (transformers help). Graph Neural Networks helps for the molecule steps. No biology background needed; step 1 covers it.
-
8 min
Step 1Protein structure and folding
The biology you need: amino acids fold into 3D shapes.
Protein Structure and FoldingAmoeba Sisters
-
8 min
Step 2The protein folding problem
Why predicting shape from sequence was a 50-year puzzle.
What is the Protein Folding Problem?The Sheekey Science Show
-
25 min
Step 3The AlphaFold story
How DeepMind cracked protein folding and why it matters.
AlphaFold - The Most Useful Thing AI Has Ever DoneVeritasium
-
13 min
Step 4How AlphaFold works
Evolution, attention, and geometry inside the model.
how AlphaFold *actually* worksLooking Glass Universe
-
10 min
Step 5Predict a structure with ColabFold
Fold your own protein for free in a notebook.
Faster AlphaFold protein structure predictions using ColabFoldUCSF ChimeraX
sokrypton/ColabFold★ 3k -
34 min
Step 6Protein language models
Language model ideas applied to amino acid sequences.
Protein Language ModelsRosettaCommons
-
44 min
Step 7Design new proteins with RFdiffusion
Diffusion models generate proteins that never existed.
RFDiffusion: Accurate protein design using structure prediction and diffusion generative modelsML for protein engineering seminar series
RosettaCommons/RFdiffusion★ 3.1k -
14 min
Step 8AI in drug discovery
Where AI helps find and test new medicines.
Revolutionizing drug discovery with artificial intelligenceRAZOR Science Show
-
17 min
Step 9Molecules in Python with RDKit
Read SMILES, draw molecules, and compute properties.
Chemistry with Python - an Introduction to RDKitBernhard Wannasek
rdkit/rdkit★ 3.7k -
16 min
Step 10Molecular property prediction with GNNs
Treat molecules as graphs to predict how they behave.
Molecular Property Prediction with Graph Neural NetworksCenter for Computer-Assisted Synthesis
deepchem/deepchem★ 7.2k -
59 min
Step 11AI weather forecasting with GraphCast
A graph network that beats classic 10-day forecasts.
GraphCast: Learning skillful medium-range global weather forecastingGreg Bronevetsky
google-deepmind/weathernext★ 7.8k -
1h 04m
Step 12Accelerating science with AI
Demis Hassabis on what AI can do for science next.
Accelerating Scientific Discovery with AI - lecture by Sir Demis HassabisCambridgeComputerLab
Build it
Watching isn't learning. Pick one and put it on GitHub.
- Fold a small protein with ColabFold and compare it with its experimental structure from the PDB.
- Use RDKit to compute properties for 100 drug molecules from SMILES and plot which ones pass Lipinski's rule of five.
- Train a DeepChem graph model to predict solubility on the ESOL dataset and report its error on a held-out split.