How soedinglab/pmgen works
Project type and purpose
• PMGen is not a web app or database-backed service, but a command-line, bioinformatics workflow/pipeline.
• Purpose: orchestrate three major sub-tools—PANDORA (template generation), AlphaFold (via AFfine, structural prediction & relaxation), and ProteinMPNN (sequence design)—to model pMHC complexes, predict binding, and optionally generate novel sequences.
File structure
• Top-level scripts:
• run_PMGen.py (main CLI entrypoint)
• run_utils.py, run_PMGen.py launch either “modeling” (single) or “wrapper” (batch) workflows.
• user_setting.py (paths to external tools, user config).
• Major folders:
• AFFine/ — contains a fork of AlphaFold (data pipelines, models, relax modules).
• PANDORA/ — contains the PANDORA template-generation tool in PANDORA/PANDORA.
• utils/ — shared Python utilities: sequence parsing, alignment, data processing, visualization.
• data/ — input examples, HLA allele databases, test TSVs.
• Config/deployment: install.sh, environment.yml, pip_requirements.txt, dockerfile—configure Conda env and clone submodules.
README for additional insights
• Describes three sequential stages:
a. PANDORA templates → b. AlphaFold modeling (+relax) → c. ProteinMPNN design.
• CLI flags control which stage(s) to run, resource limits, number of sequences, and external dependencies (Modeller, netMHCpan, GPU).
Elements
• Blue for core modules (PANDORA, AFfine, ProteinMPNN)
• Green for orchestrator & CLI
• Orange for input/output data stores
• Rectangles for processes/modules
• Clouds for external services