[chimerax-users] ChimeraX AlphaFold2

[Tom Goddard]

ChimeraX does not yet allow running AlphaFold-Multimer the version of AlphaFold released November 2, 2021 that can predict structures for molecular assemblies.  We want to offer this, although a ChimeraX version will only be a slight convenience over running an AlphaFold-Multimer Google Colab notebook with a web browser without ChimeraX.

The ChimeraX Google Colab AlphaFold predictions are limited in how long a sequence they can handle to about 800-1000 amino acids.   This  is because the old Google Colab GPUs (Nvidia K80 and T4) have only 16 Gbytes of memory and AlphaFold runs out of memory.  When you predict a molecular assembly with AlphaFold-Multimer, the sum of all sequence lengths will have the same limit.  So only small assemblies will be possible.  To predict longer sequences or larger assemblies you need a GPU with more memory such as an Nvidia A100 with 80 Gbytes, or an A40 with 48 Gbytes.  I don't know if Google Colab offers or will offer those in the future.  I am currently working on running AlphaFold-Multimer on the UCSF cluster with A40 GPUs to see how well it works and what sequence length it can handle.  The memory use is said to increase as the square of the sequence length.  We are looking at whether ChimeraX could run AlphaFold  predictions on the UCSF Wynton cluster but we probably do not have the budget to pay for that since it would probably cost about $100,000 per year to maintain about 16 high-end GPUs to meet the demand of ChimeraX users.  So we are still figuring out what AlphaFold services ChimeraX can offer.

	Tom

> On Nov 15, 2021, at 9:13 AM, Eric Pettersen via ChimeraX-users <chimerax-users at cgl.ucsf.edu> wrote:
> 
> Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
> 
> 
> 
>> On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users <chimerax-users at cgl.ucsf.edu <mailto:chimerax-users at cgl.ucsf.edu>> wrote:
>> 
>> Hi ChimeraX Dev
>> 
>> I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. 
>> Thank you
>> _______________________________________________
>> ChimeraX-users mailing list
>> ChimeraX-users at cgl.ucsf.edu <mailto:ChimeraX-users at cgl.ucsf.edu>
>> Manage subscription:
>> https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
> 
> _______________________________________________
> ChimeraX-users mailing list
> ChimeraX-users at cgl.ucsf.edu
> Manage subscription:
> https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://plato.cgl.ucsf.edu/pipermail/chimerax-users/attachments/20211115/0a900834/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Image 11-15-21 at 9.10 AM.jpeg
Type: image/jpeg
Size: 6231 bytes
Desc: not available
URL: <http://plato.cgl.ucsf.edu/pipermail/chimerax-users/attachments/20211115/0a900834/attachment.jpeg>