HPC_workflows

General-purpose scaffolding for running jobs on the Marvin HPC cluster (University of Bonn), reusable across projects (not tied to any single analysis).

Contents

• marvin_hpc_reference.md — full operating manual: access, partitions, queues, resource defaults, filesystems, module/conda conventions, and verified account details.
• templates/ — ready-to-copy SLURM job scripts (CPU, GPU, array jobs). Copy into a project's scripts/hpc_jobs/ and edit the command; don't edit these in place per-project.
• envs/ — declarative conda/module environment specs shared across projects.
• scripts/ — shared Python helper functions for job submission / result harvesting.

Conventions

• Code here is version-controlled; data is not. Large files, HPC outputs, and raw/processed data never belong in this repo — they go in ~/Documents/HPC_data/<project>/.
• Project-specific job scripts (that call these templates with real parameters) live in each project's own repo, e.g. ../Thymic_NK_development/scripts/hpc_jobs/.
• Every job on Marvin must set --account=ag_iei_abdullah (see reference doc §4).
• Harvested compute outputs should be moved from the session workspace into ~/Documents/HPC_data/<project>/hpc_runs/<date>_<description>/ and/or saved as Claude Science artifacts for provenance.

Remote

git remote add origin https://github.com/Eomesodermin/HPC_workflows.git (create the repo on GitHub first — see project setup notes).