PLM-Interact on bwVisu

Welcome to the PLM-interact Tutorial for bwVisu!

PLM-interact is an open-source protein language model that predicts whether two proteins can interact from their sequences. This tutorial will guide you through running PLM-interact on bwVisu. Please follow these steps carefully. Any feedback on the tutorial is welcome! Feel free to contact us!

Step 1: Get access to bwVisu

To start, get access to bwVisu via bwForCluster Helix or SDS. For more information, visit

https://www.urz.uni-heidelberg.de/en/service-catalogue/software-and-applications/bwvisu

For technical questions regarding the high performance cluster, see https://bw-support.scc.kit.edu. Feel free to contact us for support.

Step 2: Connect to bwVisu and Start Jupyter

Go to https://bwvisu.bwservices.uni-heidelberg.de/ and log in with your credentials and one-time password.

Choose Jupyter and start a new session. Now you can select the resources you need.

For the inference model in PLM-interact we need a GPU. A list of available GPUs and their specifications is available at https://wiki.bwhpc.de/e/Helix/Hardware#Compute_Nodes, or in the table below.

Screenshot

The GPU is selected by "GPU Type". The memory of each GPU Type is specified in GPU Memory per GPU (GB). For this example we select one of the A40 GPUs. Larger jobs (= longer sequences, more chains) require more memory. To access these, it is suggested to run the job directly on the Helix cluster. Feel free to contact us, if you need assistance!

Screenshot

Click on "Launch". This will bring you to a new screen showing your interactive sessions. Wait for your session to be ready, then click on "Connect to Jupyter". This brings you into a JupyterLab environment.

Step 3: Set a Working Directory and Upload Files

Now we need to define a working directory. These will contain all files necessary for the tutorial. A new directory can be created using folder icon on the top left of the file browser:

Screenshot

Input Sequences in `.csv` Format

PLM-interact reads the input sequence pairs from a .csv file. More information can be found here. We also provide an example in our github.

query	text
{sequence1}	{sequence2}

If you copy these from .fasta files, make sure that there are no spaces within the sequences.

Upload the PLM-interact notebook from our github and the .csv file by clicking on the upload button:

Screenshot

After the upload, you can see the notebooks in the file browser on the left.

Screenshot

Step 4: Open the Notebook and Start the Calculation

Open PLM_interact.ipynb and execute the cells in the notebook to start your Boltzgen run!

Verify Input

Before starting your Boltz prediction you should see the following files in your working directory:

Screenshot

Verify Output

In the output directory, there will be a .csv file with the output scores. For more information on the scoring, see the PLM-interact publication.

Screenshot

Note that the entries in the output .csv file are in the same order as the input sequence pairs. In the tutorial .ipynb we include a simple cell to merge the input sequences and output scores, if you prefer this:

Screenshot