Ocean Genomics

Ocean Genomics

Agentic bioinformatics skill for ocean genomics — from eDNA water sampling to species identification, protein structure prediction, and scientific insight publication.

Quick Start

1. Set Up MCP Integration

Install gget-MCP for immediate access to BLAST, AlphaFold, sequence retrieval, and 9 other genomics tools:

// Add to .claude/settings.json → mcpServers
{
  "gget-mcp": {
    "command": "uvx",
    "args": ["--from", "gget-mcp@latest", "stdio"]
  }
}

2. Install CLI Tools

# Core tools
brew install blast mmseqs2 samtools minimap2 nextflow
pip install biopython biotite gget

# Nanopore basecalling (runs on Apple Silicon)
brew install nanoporetech/dorado/dorado

# eDNA pipeline (conda recommended)
conda create -n edna -c bioconda -c conda-forge qiime2 kraken2 fastp fastqc

3. Install Companion Skills

npx skills add adaptyvbio/protein-design-skills@alphafold -g -y
npx skills add anthropics/life-sciences@nextflow-development -g -y
npx skills add gptomics/bioskills@bioskills -g -y
npx skills add davila7/claude-code-templates@gget -g -y

Core Workflows

Workflow A: eDNA Species Inventory

Trigger: User has sequencing data from marine water samples and needs species identification.

Input: FASTQ files (amplicon sequencing of CO1/16S/18S marker gene)

1. Quality control
   → fastp: trim adapters, filter low-quality reads
   → FastQC + MultiQC: generate QC reports

2. Denoise and classify
   → QIIME2 import → DADA2 denoise → ASV table
   → Taxonomy: classify-sklearn against BOLD (CO1) or SILVA (16S/18S)

3. Diversity analysis
   → Alpha diversity: Shannon, Simpson, Chao1
   → Beta diversity: Bray-Curtis, PCoA ordination
   → Rarefaction curves

4. Species report
   → Taxa count table (species × samples)
   → Novel species flags (CO1 divergence >3% from nearest reference)
   → Upload-ready format for OBIS

Output: Species inventory, diversity metrics, novel species candidates

Nextflow shortcut: nextflow run nf-core/ampliseq --input samplesheet.csv --FW_primer <fwd> --RV_primer <rev>

Workflow B: Metagenomic Community Profiling

Trigger: User has shotgun metagenomic data and needs taxonomic + functional profiles.

Input: FASTQ files (shotgun whole-metagenome sequencing)

1. QC → fastp
2. Taxonomic classification → Kraken2 + Bracken (abundance estimation)
3. Functional profiling → HUMAnN3 (metabolic pathways)
4. Assembly (optional) → SPAdes/MEGAHIT → QUAST quality check
5. Binning (optional) → MetaBAT2 → CheckM quality → MAG annotation

Output: Taxonomic profiles, functional pathways, metagenome-assembled genomes (MAGs)

Nextflow shortcut: nextflow run nf-core/mag --input samplesheet.csv --assembly_type megahit

Workflow C: Novel Protein Characterization

Trigger: User has protein sequences (from metagenomics or genome annotation) and needs structure/function prediction.

Input: Protein sequences (FASTA)

1. Homology search
   → BLAST/MMseqs2 against UniProt, GenBank nr
   → If match found: annotate function from homolog

2. Structure prediction (if novel / no homolog)
   → Fast screening: ESMFold2 single-sequence mode (no MSA) — legacy: ESMFold
   → With ligands / DNA / RNA / modified AA: AlphaFold 3 (leads both subsets); ESMFold2 if MIT required
   → High accuracy: benchmark ESMFold2 vs AlphaFold 3 / ColabFold on your targets
   → Check confidence: pLDDT >70 = reliable fold

3. Structural search
   → Foldseek against AlphaFold DB + PDB
   → Find structural homologs even without sequence similarity

4. Functional annotation
   → Domain analysis: InterPro scan
   → Pathway mapping: gget enrichr
   → Interaction networks: STRING database

5. Report
   → Structure visualization (PDB files)
   → Functional predictions with confidence scores
   → Comparison to known extremophile enzymes

Output: 3D structures, functional annotations, drug/enzyme candidates

Workflow D: Variant Effect Analysis

Trigger: User wants to understand impact of mutations in marine organism genomes.

Input: DNA sequences + variants of interest

1. Score variants with Evo 2
   → Zero-shot pathogenicity prediction
   → No fine-tuning needed — works across all species

2. Structural impact (if protein-coding)
   → Translate → AlphaFold → compare wild-type vs mutant structure
   → Stability prediction (ΔΔG estimation)

3. Conservation analysis
   → Multiple sequence alignment (MUSCLE/MAFFT)
   → Conservation scores across species

Output: Variant effect scores, structural impact, conservation context

Tool Reference

Sequence Analysis

eDNA & Taxonomy

Structure Prediction

Database Queries (via gget)

import gget

gget.blast("ATCGATCG...")          # BLAST search
gget.alphafold("MVLSPADKTNVK...")  # Structure prediction
gget.seq("ENSG00000012048")        # Fetch sequence
gget.info("BRCA1")                 # Gene metadata
gget.enrichr(["BRCA1", "TP53"])    # Pathway enrichment
gget.pdb("1BNA")                   # PDB structure

File Format Reference

Marine Databases

Foundation Models

ESM stewardship: Meta FAIR → EvolutionaryScale → Chan Zuckerberg Biohub. Repo moved to github.com/Biohub/esm and relicensed to MIT (2026-05-27); facebookresearch/esm is archived.

Accuracy is task-dependent, vendor-self-reported, and not independently replicated as of 2026-07-21: ESMFold2 beats AlphaFold 3 head-to-head on protein-protein (76 vs 73) and antibody-antigen (53 vs 47), but AF3 leads protein-ligand and protein-DNA. Benchmark your own targets.

⚠️ Local ESMFold2 needs the Biohub transformers fork (stock transformers has no esmfold2 module) and Python 3.12 only — install via pip install esm@git+https://github.com/Biohub/esm.git@main. ESMC context window: 2048 tokens — chunk longer metagenomic ORFs.

For detailed specifications on each model, see references/foundation-models.md.

Detailed References

• references/tools-inventory.md — Full CLI commands, install instructions, and usage examples for all 30+ tools
• references/marine-databases.md — Database catalog with API access patterns
• references/foundation-models.md — Technical specs, hardware requirements, and code examples for each model
• references/mcp-integration.md — MCP server setup, agent workflow architecture, and skill companion catalog