Skill: Search NCBI Datasets

Use When

• A user wants to find and download genome assemblies from the NCBI Datasets portal.
• A user needs to query genomes by organism name, taxonomy ID, assembly accession (GCF/GCA), or assembly level (complete, chromosome, scaffold, contig).
• A user wants gene-level data: sequences, annotations, or orthologs by gene symbol, gene ID, or taxon.
• A user wants virus genome records (NCBI Virus) by species, collection date, host, or geographic region.
• A user wants summary statistics (assembly count, N50, total length) for a set of assemblies before downloading.
• A user wants to generate a dehydrated dataset package and rehydrate later, or get direct download links for genome FASTA/GFF/GTF files.

Inputs

• Required:
  • one of: organism name or NCBI taxonomy ID, assembly accession (GCF_/GCA_), gene symbol or gene ID, or virus taxon
• Optional:
  • data type: genome, gene, virus, taxonomy (default: genome)
  • assembly level filter: complete, chromosome, scaffold, contig
  • assembly source filter: refseq (GCF) or genbank (GCA)
  • annotation filter: annotated (only assemblies with annotation)
  • reference-only flag: restrict to reference/representative genomes
  • include flags for genome downloads: genome, rna, protein, cds, gff3, gtf, seq-report
  • limit on number of results
  • output format: json (summary) or direct download (zip package)
  • output directory

Workflow

1. Choose the interface:
   • NCBI Datasets CLI (datasets + dataformat): preferred for batch or scripted use.
   • NCBI Datasets REST API: use when the CLI is not installed. Base URL: https://api.ncbi.nlm.nih.gov/datasets/v2/
2. For genome assembly searches:
   • CLI: datasets summary genome taxon "<organism>" --assembly-level complete --reference
   • API: GET https://api.ncbi.nlm.nih.gov/datasets/v2/genome/taxon/<taxon>/dataset_report
   • Reformat summary with: dataformat tsv genome --fields accession,organism-name,assembly-level,genome-size,contig-n50,annotation-info-release-date
3. For gene searches:
   • CLI: datasets summary gene symbol <GENE> --taxon "<organism>"
   • CLI: datasets summary gene id <gene_id>
4. For virus genome searches:
   • CLI: datasets summary virus genome taxon <taxon> --host "<host>" --geo-location "<country>"
   • Filter by collection date range with --released-after / --released-before.
5. To download data:
   • datasets download genome accession <GCF_> --include genome,gff3,gtf --filename <out>.zip
   • For large batches, use --dehydrated and then datasets rehydrate --directory <dir>.
6. Parse the JSON summary to extract key fields and report a summary table.

Output Contract

• Exact CLI command or API URL used
• Number of matching records
• Summary table for genome queries: Accession (GCF/GCA), Organism, Assembly Level, Genome Size (Mb), Contig N50, Annotation Date, Submitter
• Summary table for gene queries: Gene ID, Symbol, Full Name, Taxon, Location, RefSeq mRNA/Protein accessions
• Download command(s) for selected accessions
• Saved output files when a directory is provided:
  • summary.json (raw API/CLI output)
  • summary.tsv (reformatted with dataformat)
  • accessions.txt (one accession per line)

Limits

• This skill depends on live access to https://api.ncbi.nlm.nih.gov/datasets/v2/ or a local install of the datasets CLI.
• NCBI Datasets API rate limits apply; add NCBI_API_KEY environment variable or --api-key flag for higher throughput.
• The datasets CLI must be installed separately (available from https://www.ncbi.nlm.nih.gov/datasets/docs/v2/download-and-install/).
• Genome download packages can be very large; always check genome size and count before issuing a download.
• Not all assemblies in GenBank have annotation; use --annotated when annotation files are required.
• Virus genome data is stored separately from the main genome database; queries must use datasets summary virus.
• Common failure cases:
  • specifying a taxon name with incorrect spelling or outdated taxonomy
  • requesting --include gff3 for assemblies without annotation (download will still proceed but GFF3 will be absent)
  • confusing GCF (RefSeq) with GCA (GenBank) accessions when a specific source is required
  • omitting dataformat to reformat JSON output, resulting in hard-to-parse raw JSON