alphafold-notebook-demo

Demonstration workflow with Alphafold in a Jupyter notebook

Setup

The following components are necessary for setting up this workflow:

1. An Alphafold Singularity container. Please see instructions in ./container for how to build an Alphafold container. Currently, it is assumed that this container is available at a hard coded path in ./container/run_singularity_container.py in this line of code:

singularity_image = Client.load('/public/apps/alphafold/alphafold.sif')

2. A Conda (or pip) environment that has the absl-py and spython packages to launch the container. This workflow also uses parsl (but it is not required for using the container itself). For a cluster with Conda in a module, here is an example for how to create a local environment:

module load conda3
source /gs/gsfs0/hpc01/rhel8/apps/conda3/etc/profile.d/conda.sh
conda create -y -p /gs/gsfs0/users/gstefan/work/alphafold/env -c conda-forge absl-py==0.13.0 spython=0.1.16 parsl
conda activate /gs/gsfs0/users/gstefan/work/alphafold/env

where /gs/gsfs0/users/gstefan/ is your home directory.

3. Pull this workflow code into your PW environment. TODO: Add instructions here.

4. Run the workflow from PW. TODO: Add instructions here.

Interactive runs

For the purposes of testing Alphafold, it is possible to start interactive Alphafold runs (i.e. manually launch the application for an instance). Instructions for launching an interactive run are in ./container.

Batch runs

When you want to run many proteins with Alphafold, there are two options:

1. the workflow form (under construction) can be used to launch a batch run or
2. main.ipynb, the Jupyter notebook that contains the workflow code.

When users opt for the first option (the workflow form), the form simply grabs the code out of main.ipynb and executes it. Users can use main.ipynb as a template for more complicated Alphafold workflows and/or directly modify some of the Alphafold options that are not available in the workflow form. Jupyter notebooks (*.ipynb files) can be opened, edited, and run on the platform by double clicking on the file in the file browser pane.

ColabFold

ColabFold is a community-driven update to Alphafold underpinned by new/updated databases and the MSA search process is accelerated by MMseqs2. Please see the colabfold directory for more information.