Skip to content

outputs.md

Latest commit

 

History

History
102 lines (89 loc) · 5.61 KB

outputs.md

File metadata and controls

102 lines (89 loc) · 5.61 KB
Raw

Output reference

Folder layout

04_output/
├── final_annotations_table.csv     # deliverable table
├── final_annotations_table.xlsx    # colour-coded
├── updated_prophages.gb            # curated GenBank (not produced in pangenome sub-mode)
├── pangenome_reference.csv         # pangenome sub-mode only: one row per group
├── pangenome_reference.xlsx        #   (see below)
├── pangenome_reference.faa         #   DIAMOND/MMseqs-ready reference FASTA
└── curation/
    ├── curated_annotations.csv     # written by 04_curate_annotations.py
    └── review_suggested.csv        # written by 05_build_output.py
                                     # (both run in one job, steps/04_curate.sh)
05_phynteny/
├── phynteny.tsv  phynteny.gbk  phynteny_predictions.csv
├── final_annotations_integrated.csv  # phynteny merged into final_function
├── final_with_synteny.gb             # GBK with synteny/phynteny notes
└── run/
    ├── phynteny_run.log            # phynteny_transformer's verbose log
    └── phynteny_input.fasta        # the FASTA phynteny wrote (relocated here)

04_output/final_annotations_table.{csv,xlsx}

One row per CDS. XLSX is colour-coded (resolved = green, still-hypothetical = orange). Key columns:

Identity & answer

  • prophage, locus_tag, aa_length
  • short_name — concise label (see naming rules)
  • final_product, final_function — the curated call
  • was_hypothetical — was this a Pharokka hypothetical (i.e. a pipeline target)?
  • annotation_source — evidence category (see below)
  • best_hit_kingdom — Bacteria/Archaea/Virus/Eukaryote/Unknown

Pharokka: pharokka_product, pharokka_function Phold: phold_product, phold_function_cat, phold_confidence, phold_phrog, accession_phrog, phold_evalue Curation: agreement (from step 03, carried verbatim), curation_action (which rule resolved the gene), curation_suggestion / curation_explanation (the rule engine's proposed call and its stated reasoning — always deterministic, rule-based; no LLM/AI is involved in generating these) Custom FoldSeek: foldseek_description, foldseek_accession, foldseek_db, foldseek_taxname, foldseek_confidence, foldseek_score, foldseek_evalue, foldseek_pident, foldseek_qcov_frac, foldseek_partial_match, foldseek_same_host, foldseek_top3 (a.k.a. fs_top3), fs_top3_kingdoms — the kingdom (Bacteria/Archaea/Viruses/Eukaryota/Unknown) of each of the FoldSeek top-3 hits, positionally parallel to fs_top3 Provenance: note (pipeline=phagefactor; source=…; …; original_pharokka=hypothetical protein)

Evidence categories (annotation_source)

Value Meaning
both agree Phold and custom-FoldSeek concur (strong/partial)
merged combined into one more-specific call
foldseek → structural-only FoldSeek structural homology, no Phold corroboration
phold → phold-only Phold call, no FoldSeek support
pharokka non-hypothetical, kept from Pharokka
manual_review / curation_suggestion needs_review rows (reviewer-edited / rule engine's suggestion accepted as-is)
structural-uncharacterised a CONFIDENT/GOOD fold match to an unnamed target (written by 07_integrate). final_product keeps the target's header (often a DUF), so the fold is recorded — but the FUNCTION is unknown. Do not count these as annotated.
no_hit still hypothetical

Reader-facing rename: surface foldseek as structural-only in figures/tables.

04_output/updated_prophages.gb

GenBank with curated /product, /function, and /note provenance per CDS. Built by one of three paths (update Pharokka GBK / concat per-prophage GBKs / build from coords for the pre-CDS sub-mode). Not produced in pangenome sub-mode (no genome coordinates).

04_output/pangenome_reference.{csv,xlsx,faa} (pangenome sub-mode)

Built by steps/08_pangenome_reference.sh after curation, when an id map (input/pangenome_id_map.csv) is present. Rejoins the final table (keyed on the clean pang_NNNNN ids) back to the biological group names + sequences.

  • .csv / .xlsx — one row per pan-genome group: group_id, representative_locus, clean_id, aa_length, aa_sequence, final_product, short_name, annotation_source, agreement (plus any evidence columns present).
  • .faa — the same representatives as a FASTA, header >group_id final_product. Ready for diamond makedb / mmseqs createdb; a homology hit's sseqid is the group_id, so annotation transfer to a larger protein set is a plain join back to the CSV.

Phynteny step (05_phynteny/)

  • 05_phynteny/phynteny_predictions.csv — synteny category + probability (≥0.8).
  • 05_phynteny/final_annotations_integrated.csv — phynteny category merged into final_function (phynteny priority on confident gaps) + synteny_hint (C1/Cro, lysis cassette, integration module).
  • 05_phynteny/final_with_synteny.gb — GBK with synteny/phynteny notes.
  • 05_phynteny/run/ — phynteny_run.log (phynteny_transformer's verbose log) and phynteny_input.fasta (the FASTA it wrote), relocated/renamed there by steps/05_phynteny.sh.

04_output/curation/review_suggested.csv

Genes flagged for manual review (written by 05_build_output.py, part of steps/04_curate.sh). Read curation_explanation for the rule engine's stated reasoning, accept/edit final_annotation, then re-run. Pre-filled with curation_suggestion. (04_curate_annotations.py, the other half of the same job, writes the companion 04_output/curation/curated_annotations.csv.)