bio-gene-calling

Bio Gene Calling

Call genes and annotate basic features for prokaryotes, viruses, and eukaryotes.

Instructions

Tool guides and versions: docs/README.md.

1. Validate the assembly and tool manifests, then generate a per-assembly execution plan:

   uv run --script skills/bio-gene-calling/scripts/run_gene_calling.py \
     assemblies.tsv --tool-manifest tool-manifest.json \
     --out results/bio-gene-calling
   # Inspect run_manifest.json, then execute or resume the same plan:
   uv run --no-project python skills/bio-gene-calling/scripts/run_gene_calling.py \
     assemblies.tsv --tool-manifest tool-manifest.json \
     --out results/bio-gene-calling --execute
   

   The tool manifest must pin the BRAKER4 repository commit, Snakefile checksum, container-lock checksum, and every Rfam model checksum. The driver records input FASTA checksums, routes each assembly by domain, writes idempotent BRAKER4 samples.csv and config.ini files, uses absolute covariance-model paths, and materializes the required default/relaxed ncRNA census rows. Execution reuses only non-empty declared outputs and replaces pending ncRNA counts with parsed tRNAscan-SE and Infernal counts.

2. Select gene caller by organism class:

   • Bacteria and Archaea: Pyrodigal v3.7+ with single-genome or metagenomic mode chosen from the input.
   • Viruses, including giant and alternative-code viruses: pyrodigal-gv v0.3+ with the appropriate viral model.
   • Eukaryotes: BRAKER4 as the current upstream workflow. Pin the tested repository commit, workflow checksum, and container lock in provenance. BRAKER4 is a Snakemake workflow; do not invoke the legacy braker.pl entry point for a BRAKER4 run. Keep BRAKER3 only for a documented legacy reproduction.

3. For eukaryotic/protist drafts with ONT cDNA or other transcriptome reads, build a transcript evidence bundle before gene calling:

   • Orient/filter full-length ONT cDNA reads with the Pychopper guidance in /bio-reads-qc-mapping, including plain .fastq output handling and resume from existing classified reads after report-plotting failures.
   • Map transcript reads splice-aware to each candidate draft genome with minimap2 (-ax splice family settings appropriate to the organism/data), sort/index BAMs, and compute a per-genome mapped fraction table.
   • Use the best-supported draft genome as the primary evidence target, but keep the full mapping table because it documents sample/genome assignment and cross-sample ambiguity.
   • Produce StringTie long-read GTF/transcript FASTA and, when useful, a reference-free transcript assembly such as RNA-Bloom. Summarize these paths in a gene_calling_evidence.tsv bundle with columns: sample_id, evidence_type, genome_id, path, notes. The bundle should be directly usable by BRAKER4 or another eukaryote-aware caller.

4. Run gene calling and produce per-assembly GFF/protein/CDS outputs. BRAKER4 emits compressed results under output/{sample}/results/; Pyrodigal and pyrodigal-gv emit the normalized uncompressed paths recorded in the run manifest.

5. Always run tRNA detection and rRNA detection on every assembly, and report counts per class. Negative findings (zero hits at default and relaxed thresholds) are required results — never leave ncRNA presence/absence unstated.

   • tRNA: tRNAscan-SE v2.0.12+ (preferred; isotype-specific covariance models) or ARAGORN v1.2.41+ for tmRNA where appropriate.
   • rRNA: Infernal v1.1.5+ cmsearch against the relevant Rfam covariance models. Pick the model set by domain of life:
     • Bacteria: RF00177 (SSU 16S), RF02541 (LSU 23S), RF00001 (5S).
     • Archaea: RF01959 (SSU 16S), RF02540 (LSU 23S), RF00001 (5S).
     • Eukaryotes: RF01960 (SSU 18S), RF02543 (LSU 28S), RF00002 (5.8S), RF00001 (5S).
     • Metazoan mitochondria, when applicable: RF02555 (12S), RF02546 (16S). cmsearch --rfam --cut_ga --nohmmonly is a sensible default; if no hits, rerun without --cut_ga and record both results.

6. For viral or otherwise specialized genomes, choose the gene caller and mode from tool documentation and the literature-derived analysis playbook for the inferred group; record the rationale.

7. Summarize gene count, gene density, coding fraction, ORF length distribution, unusually long ORFs, overlapping genes, tRNAs, rRNAs, and other features that may affect downstream discovery.

8. Flag gene-calling anomalies relative to the inferred group and data type, including patterns that could hide interesting biology or indicate artifacts.

9. Produce a ncRNA_census.tsv with columns: assembly, class (tRNA/rRNA/tmRNA/other), tool, model (Rfam accession when applicable), threshold (default/relaxed), count, notes. This file is required even when all counts are zero.