Reproducibility

Reproducibility

CheckAMG is packaged with curated reference tables and pre-built databases. To keep these resources transparent and reproducible, the repository includes the notebooks used to build them and to train, evaluate, and benchmark the models behind CheckAMG. These notebooks are not required to run CheckAMG; they are provided so others can inspect, regenerate, and update the curated assets and reproduce the manuscript analyses.

All notebooks live under notebooks/. Figures for the manuscript are generated from the paired R Markdown (.Rmd) files rather than the notebooks themselves; each analysis notebook exports tables that its corresponding .Rmd renders.

Some notebooks are released alongside the manuscript and may be added incrementally. Entries below whose links do not yet resolve are placeholders pending the manuscript release.

Curated reference tables and databases

CheckAMG reads several curated tables under CheckAMG/files/ that define (i) the functional label mappings used for reporting, (ii) the curated AMG/APG/AReG HMMs, and (iii) the AVG filtering tables (including exception categories and AMG weights).

Notebook	What it does
make_checkamg_required_tables.ipynb	Builds the required reference tables from upstream sources: hmm_id_to_name.tsv (cross-database HMM id to name/description mapping), FOAM.tsv, vscores.tsv, AMGs.tsv/APGs.tsv/AReGs.tsv, AMG_filters.tsv/APG_filters.tsv/AReG_filters.tsv, viral_hallmark_genes.tsv, and mobile_genes.tsv. Also documents the AMG weight derivation.
build_annotate_db.ipynb	Downloads, formats, and packages the standardized annotate HMM database (KEGG, FOAM, Pfam, PHROGs, dbCAN, METABOLIC, CAMPER) with database-provided cutoffs. Documents source versions.
build_denovo_db.ipynb	Assembles the de-novo database: the released CheckAMG-PST model checkpoint, reference protein PST embeddings, FAISS index, and labels.

More under construction