We updated our bioRxiv preprint last week to now include a succinct dimensional analysis of the major denoising diffusion equations. We summarize with a table that shows the meaning of each term in both AI and physics jargon.

Our goal in this project was to explore methods whereby one might extract some ‘potential’ from AlphaFold3 to see why it is so much better than our traditional physics based approaches.

@britnie-carpentier.bsky.social presenting her Rosetta energy approximation network at European RosettaCon!

In these pictures, we are at the historic castle of Bled…lab members Britnie, Natalia, Fatima and friend Rebecca are the antibody and I’m an antigen! Thanks to @ajasjaljubetic.bsky.social and Dominic Gront for being amazing conference chairs and hosts!

A comparison of DFMDock performance to Rosetta's top ranked structure. Read the preprint here: www.biorxiv.org/content/10.1...

By using an additional prediction head for an energy with its gradient matched to the score, we extract an energy function and compare it to Rosetta, an empirically tuned physically-inspired energy function. The learned energy sometimes recovers funnels with near-native minima, and sometimes not!

Would love to follow jazzway here instead of... Anywhere else