Skip to content

Monthly Deep Research Update Prompt (standard)

Use this prompt verbatim in the Deep Research run each month. Results go to update/candidates/YYYY-MM/*.yaml as draft tool-registry entries.

System / role

You are scouting the bioinformatics literature and tool ecosystem for the bioflow registry. Focus on peer-reviewed papers and high-quality preprints published in the last 60 days (cover two months to avoid gaps).

Pipelines & stages in scope

  1. Genome Assembly & Annotation
  2. step1 Read QC (short / HiFi / ONT)
  3. step2 Assembly (de novo short, de novo HiFi, de novo ONT, hybrid, resequencing consensus)
  4. step3 Assembly QC (contiguity, completeness, base accuracy)
  5. step4 Repeat masking (eukaryote)
  6. step5 Structural annotation (prokaryote, eukaryote)

  7. step6 Functional annotation

  8. RNA-seq DEG

  9. step1 RNA-seq QC
  10. step2 Alignment / alignment-free quantification
  11. step3 Differential expression (DESeq2/edgeR/limma etc.)
  12. step4 GO / KEGG / GSEA enrichment

For each newly-recommended tool, return

  • id (lowercase, snake_case)
  • name, version, citation (DOI or PMID)
  • which category and stage it belongs to
  • applicable: species, read_type, mode
  • container.image — prefer BioContainers / quay.io / staphb / community
  • resources.min and resources.recommended (cpu, ram_gb, disk_gb)
  • gpu required? arch list
  • A one-line command_template
  • Benchmark summary vs current registry incumbent (speed, accuracy, memory) — cite numbers from the paper, do not estimate
  • Risks: licensing, maintenance status (last commit), open issues

Acceptance criteria

Only recommend a tool for promotion to the registry if it meets ALL:

  1. Peer-reviewed or strong preprint with benchmark vs established tools
  2. Has a publicly pullable container image, OR a trivially-buildable Dockerfile
  3. Outperforms (or is a viable alternative to) the current registry pick on at least one meaningful axis (speed, accuracy, memory, ease-of-use)
  4. No paywalled reference DB is strictly required (or a free mirror exists)

Output format

YAML per tool, one file per candidate, matching registry/schema.yaml.

Add a top-level update_meta: block per file with:

update_meta:
  month: YYYY-MM
  replaces: [existing_tool_id, ...]   # or []
  benchmark_note: "..."
  risks: ["..."]