Skip to content

public-databases.md

Latest commit

 

History

History
74 lines (61 loc) · 3.43 KB

public-databases.md

File metadata and controls

74 lines (61 loc) · 3.43 KB
Raw

Public databases

Public metagenomics resources are a data skill layer separate from analysis tools. Start at skills/public-databases/SKILL.md, then open the per-resource skill.

Resource → skill

Resource Skill Typical use
NCBI SRA / EBI ENA sra-ena Download public shotgun reads
MGnify (+ UHGG / biome catalogues) mgnify Analysed studies; MAG/protein catalogues
GTDB releases (e.g. R10-RS226) gtdb Taxonomy tables, trees, genome reps
proGenomes4 progenomes ~2M QC'd annotated prokaryotic genomes
CAMI / CAMI II (+ CAMISIM, AMBER, OPAL) cami Gold-standard benchmarks & evaluation
KEGG (KO / pathways; license applies) kegg Orthology & pathway reference
Virtual Metabolic Human vmh Human/gut metabolism knowledge for GEMs
InSilicoSeq insilicoseq Illumina metagenome read simulation
OMG / gLM2 omg Metagenomic LM pretraining corpus + mixed-modality gLM
ChocoPhlAn / UniRef / MetaCyc biobakery-databases MetaPhlAn / HUMAnN indexes
Kraken2 indexes kraken2 k-mer classification DBs
eggNOG eggnog-mapper Orthology DB for annotation
CARD rgi AMR reference
CheckM2 DB checkm2 MAG quality models
Host Bowtie2 indexes kneaddata Host depletion
BASALT weights basalt / setup Refinement models

Four data layers (do not mix unlabeled)

1. Raw archives          SRA / ENA
2. Genome catalogues     MGnify biomes, GTDB reps, proGenomes
3. Tool indexes          Kraken, ChocoPhlAn, CheckM2, CARD, GTDB-Tk data
4. Benchmarks            CAMI / CAMI II gold standards (CAMISIM simulations)

Quick chooser

You need… Open
Reanalyse a published shotgun study sra-ena (± mgnify for analysed context)
Compare your MAGs to human-gut species mgnify (UHGG) + gtdbtk
Cite / update prokaryotic taxonomy gtdb (pin release, e.g. R10)
Build a large annotated reference genome set progenomes
Benchmark a binner/profiler with truth cami
KO / KEGG pathway maps kegg (confirm access rights)
Marker + pathway profiling indexes biobakery-databases

Thinking checklist

  1. Separate raw archives, catalogues, tool indexes, and benchmarks
  2. Pin release / accession / dataset DOI / download date (checksum when available)
  3. Do not merge MetaPhlAn, Kraken, and GTDB labels into one table
  4. UHGG/MGnify membership ≠ detection in your sample without abundance evidence
  5. CAMI “best tool” is dataset-conditioned — report which challenge set you used
  6. Sequence abundance ≠ taxonomic abundance in profiler benchmarks

AI / language-model & structure-search data

For genome language models start at genome-language-model (omg, nucleotide-transformer, dnabert2, caduceus, evo2). For protein language models start at protein-language-model (esm, esmc, saprot, prostt5, plmsearch, deepblast). For AFDB/PDB/ESM Atlas structure indexes use foldseek / folddisco. For vector-DB homology use erast; for phylogeny-compressed genome search collections use phylign (see also homology-search). These are ML corpora / structure / search indexes, not substitutes for SRA reanalysis or GTDB taxonomy claims.

See also