Hi David

This observation is observed by many. Nothing published though, just many observations published online on Twitter, Bluesky, here etc. Mostly imo due to the structures being diffused now. But what you get is a speed and ligands.

for most I think it’s fine. For those in the know, it’s better to just do a very quick sanity check. For use in model building though, the AF2 model I find is usually better. 

Sent from Outlook for iOS
From: David S. Fay via ChimeraX-users <chimerax-users@cgl.ucsf.edu>
Sent: Sunday, June 8, 2025 6:14:12 PM
To: ChimeraX Users Help <chimerax-users@cgl.ucsf.edu>
Subject: [chimerax-users] ChimeraX adjacent question about AF2 vs AF3
 
Greetings,

I am curious if others have found AlphaFold3 to be slightly worse or less consistent at predicting protein-protein interactions than AlphaFold2. From the AF3 paper it appeared that they were roughly equivalent, with AF3 perhaps having a slight edge. Certainly when it comes to speed and the handling of large proteins, AF3 is far superior. But over time I’ve found a number of cases where AF2 did a better job based on experimentally supported evidence from our lab. Admittedly, this observation is entirely anecdotal and likely prone to bias. And I get that AF2 may predict some interactions better than AF3 and vice versa. But it’s still makes me wonder and I was curious if others had noticed this pattern or had thoughts on the matter.

Thanks in advance.

David

_______________________________________________
ChimeraX-users mailing list -- chimerax-users@cgl.ucsf.edu
To unsubscribe send an email to chimerax-users-leave@cgl.ucsf.edu
Archives: https://mail.cgl.ucsf.edu/mailman/archives/list/chimerax-users@cgl.ucsf.edu/