AlphaFold 3¶
Operating System:
Terminal:
Shell:
Editor:
Package Manager:
Programming Language:
Database:
Extension:
AlphaFold 3 is an AI model developed by Google DeepMind and Isomorphic Labs which accurately predicts the structure of proteins, DNA, RNA, ligands and more, and how they interact.
On UCloud, AlphaFold 3 is deployed in three different modes to suit various research needs: Batch, Lab, Visualization.
General Options¶
Initialization¶
For information on how to use the Initialization parameter, please refer to the Initialization: Bash Script, Initialization: Conda Packages, and Initialization: PyPI Packages section of the documentation.
Genetic databases¶
AlphaFold 3 needs multiple genetic (sequence) protein and RNA databases to run. The necessary databases are by default mounted when the job start.
It is possible instead, to specify a custom folder containing the databases (e.g. previously downloaded by running the script fetch_databases.sh in the app terminal), by using the Personal database directory parameter.
In both cases the databases will be available in the directory /home/ucloud/public_databases.
Model parameters¶
AlphaFold 3 model parameters can be used only if received directly from Google. The instructions on how to request the access can be found here.
Once the model parameters directory is available, it can be stored on UCloud and selected using the Model parameters dir option.
AlphaFold 3 input¶
Inputs can be provided in two different way:
As a single JSON input file, through the Input JSON file option
As multiple input files through the Input directory option, which allows to select the path to a directory containing JSON files.
An example of JSON input file is the following:
{
"name": "2PV7",
"sequences": [
{
"protein": {
"id": ["A", "B"],
"sequence": "GMRESYANENQFGFKTINSDIHKIVIVGGYGKLGGLFARYLRASGYPISILDREDWAVAESILANADVVIVSVPINLTLETIERLKPYLTENMLLADLTSVKREPLAKMLEVHTGAVLGLHPMFGADIASMAKQVVVRCDGRFPERYEWLLEQIQIWGAKIYQTNATEHDHNMTYIQALRHFSTFANGLHLSKQPINLANLLALSSPIYRLELAMIGRLFAQDAELYADIIMDKSENLAVIETLKQTYDEALTFFENNDRQGFIDAFHKVRDWFGDYSEQFLKESRQLLQQANDLKQG"
}
}
],
"modelSeeds": [1],
"dialect": "alphafold3",
"version": 1
}
Details and guide on how to prepare the JSON files can be found in the AlphaFold 3 docs.
Batch Mode¶
The app runs the script run_alphafold.py.
Several parameters can be provided through the app options, in the Job submission page. The full list can be obtained by running the following command on the app terminal:
$ run_alphafold.py --help
Output folder¶
In batch mode, the app requires to specify the output folder, through the Output directory mandatory parameter. This is the directory where the results of the AlphaFold 3 run will be stored.
Lab Mode¶
The app starts an interactive JupyterLab session, where AlphaFold 3 pipelines can run.
Visualization Mode¶
The app allows to start a web interface to visualize the AlphaFold 3 prediction results.
After interface is opened, three files are necessary for the structure visualization:
a model.cif file, a confidences.json file and a summary_confidences.json file.
Output folder¶
The app requires to specify the folder where the results of an AlphaFold 3 run have been stored, through the Output directory mandatory parameter. The selected folder corresponds, e.g., to the Output directory obtained by running AlphaFold 3 in batch mode.
Contents