"Comparing protein structures with Clojure"
summary
Compares an experimental crystal structure to an AlphaFold2 model, gets a conventional fit using 235 atom pairs with ~0.5 Å RMSD, then adapts a probabilistic approach (Theseus PP → Theseus-HMC) implemented by bridging Clojure to Python via libpython-clj and running PiMC and PiTensor to do Hamiltonian Monte Carlo/NUTS on the first 100 residues. Visualizations use a Clojurescript 3dmol.js wrapper and the probabilistic fit reveals per-residue variance and improved central alignment.
AI-written summary · about
chapters (20)
- 0:00Start
- 1:05Outline
- 2:05Polypeptides Fold into Proteins
- 3:52Protein Structure Defines Protein Function
- 6:11Workflow in Our Lab
- 12:29AlphaFold Workflow
- 13:20Comparison of Al and Experiment
- 14:17Experiment (green) vs Prediction (cyan)
- 15:02Little Open Space Inside Protein Structures
- 15:56Defining the Ca-trace
- 17:08Output from align
- 18:42Superposition in PyMOL with align
- 21:56Graphical Model
- 23:05Pseudocode (1/2)
- 25:44Random Walks vs Hamiltonian Monte Carlo
- 28:50Code Snippet
- 29:44Hyperparameter Tuning with NUTS
- 30:22Conclusions
- 31:13Future Directions
- 31:50Acknowledgements
related talks
38mScientific Clojure, a bird's eye view
31mScientific Computing with Clojure
34mIntro to Probabilistic Programming and Clojure’s Anglican
43mDoing Data Science with Clojure
26mDoing data science with Clojure: the ugly, the sad, the joyful
27mReal World AI Integration
30mWolframite: Bring computational intelligence to Clojure
33mClojure Data Science