How labstructbioinf/plm-blast works
Project type and purpose
• pLM-BLAST is a Python-based command-line tool and library for sensitive remote homology detection.
• It exposes a core API (the alntools and embedders packages) as well as several CLI scripts (embeddings.py, plmblast.py, plus helper scripts in scripts/) and example workflows under examples/.
• Key workflows:
• Generate protein embeddings from FASTA/CSV → store to disk (via embeddings.py, using embedders/*)
• Search a database → pre-screen via cosine similarity → perform alignment → post-process → output CSV (plmblast.py)
File structure
• Top-level: configuration (pyproject.toml, requirements.txt), CI workflow, packaging metadata.
• Core library:
• alntools/
– alignment.py, density/, parser.py, prepare/, postprocess/, filehandle.py, settings.py
• embedders/
– Base class + specific implementations (pt, esm, hfautomodel, etc.), dataset.py, checkpoint.py, schema.py
• CLI entrypoints:
• embeddings.py (create or resume embedding generation)
• plmblast.py (searching & alignment)
• Helper scripts in scripts/ (plotting, CSV formatting, merging, etc.)
• Tests under tests/ and examples under examples/
README for additional insights
• Detailed usage for database creation, checkpointing, prescreening, search, and visualization
• Dependencies: Python ≥3.9, Transformers models, optional GPU/multiprocessing
• Emphasis on modular embedders and adaptive batching