BioM3-workspace-template

Setup

Install BioM3

Source the environment before running pipeline steps:

Symlink weights

Pretrained model weights are stored in BioM3-data-share. Symlink them into your workspace:

Symlink databases (optional)

For local BLAST searches, symlink reference databases:

Build a dataset (optional)

To build a training dataset from reference databases instead of providing your own CSV:

This extracts sequences matching the Pfam family from SwissProt and Pfam databases, producing a 4-column dataset.csv. Database and training data paths are configured in configs/dbio_config.json and can be overridden with CLI flags. Optional flags add UniProt annotation enrichment (--enrich-pfam) and NCBI taxonomy lineage (--add-taxonomy). For database syncing and one-time cache setup (Parquet conversion, annotation caching, taxonomy indexing), see docs/database_setup.md. For full Step 100 usage, see docs/0100_build_dataset.md.

Add your data

Option A: Build from reference databases using Step 100 (see above). Set training_csv = "data/ /dataset.csv" in your TOML config (matching biom3_build_dataset's output filename).

Option B: Place your own CSV under data/ /. CSVs should contain columns: protein_sequence, primary_Accession, and a text description column.

See QUICKSTART.md for detailed instructions on data format, pipeline configuration, and running the pipeline.