id	summary	reporter	owner	description	type	status	priority	milestone	component	version	resolution	keywords	cc	blockedby	blocking	notify_on_close	platform	project
5043	Add an AlphaFold structure prediction tool	Tom Goddard	Tom Goddard	"Would like the user to be able to specify a sequence and compute a predicted structure using AlphaFold.

An example of a server that does this is the AlphaFold Google Colab

  https://colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb

which uses free virtual machines with GPUs provided by Google that run Python in a web browser.

There are several requirements needed to run AlphaFold that are difficult to meet on the researcher's desktop computer so the compute job needs to be run on a server.  Some of the requirements are:

 1) Nvidia GPU supporting CUDA is needed. I believe AlphaFold cannot run without this because it uses DeepMinds JAX library for just-in-time compilation to the GPU and derivative calculation.

 2) Linux operating system.  It appears that AlphaFold has not been run on Windows.  Mac does not support Nvidia GPU.

 3) Large sequence and structure databases, ~2.2 Tbytes ideally on a fast SSD drive.   The entire sequence database will be scanned for creating a multiple sequence alignment many thousands of sequences deep.  Smaller databases can be used.  The above mentioned Google Colab server uses 150 Gbytes that is streamed from the web on every prediction.

 4) Multi-hour computation for each structure prediction.  A 500 amino acid test with Google Colab took about 2 hours.  For a ChimeraX service there may be hundreds of jobs running per day, requiring several hundred GPU compute hours per day.  This amount of compute resources will be expensive to maintain.  Google Colab is free even for commercial use but service terms are subject to change, and there are no guarantees of available resources, even getting a virtual machine with GPU depends on availability.  Google offers a $10 Colab Pro service that gives priority access to resources, but again with no guarantees.  Users could be encouraged to use their own Colab Pro account if ChimeraX utilized Google Colab resources."	enhancement	assigned	moderate		Structure Analysis								all	ChimeraX