So Orb has blown me away again. I simulated the carbonic anhydrase enzyme with it: one of the most important and well studied enzymes in biology. (It converts CO2 to bicarbonate and is involved in many diseases and could also be useful for carbon capture.)

Running out of memory used to be a common headache when running molecular simulations with neural network potentials. Not any more. Here Orb is simulating over half a million atoms on a single GPU (H200). This is a fully solvated COVID spike protein.

Models here: github.com/orbital-mate...

Uranium is one of the hardest elements to simulate due to the large number of electrons, so I thought I would try it with Orb and remarkably it seems to behaves well even getting the melting point roughly correct. I think looking at this systematically for many metals would be a great project.

Here's a fun one you can't do in the lab: diamond melting at thousands of degrees simulated with Orb. Simulate anything you want with it here: colab.research.google.com/github/timdu...

Simulating over 10,000 atoms for 10 ps a day on my Macbook with close to quantum chemical accuracy using Orb. I can do this for almost any element from the periodic table I want. Just a few years ago this would have been totally inconceivable with even the world's largest supercomputers.

So I think I've found another pretty incredible example of the generalisability of neural network potentials: this is a problem I've been dreaming of tackling for a decade but never felt I had the the tools to get at until now: How do potassium ion channels work. 1/n