Run bioinformatics analyses using Lobster AI - single-cell RNA-seq, bulk RNA-seq, literature mining, dataset discovery, quality control, and visualization. Use when analyzing genomics data, searching for papers/datasets, or working with H5AD, CSV, GEO/SRA accessions, or biological data. Requires lobster-ai package installed.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Run bioinformatics analyses using Lobster AI - single-cell RNA-seq, bulk RNA-seq, literature mining, dataset discovery, quality control, and visualization. Use when analyzing genomics data, searching for papers/datasets, or working with H5AD, CSV, GEO/SRA accessions, or biological data. Requires lobster-ai package installed.
Lobster Bioinformatics Agent
Lobster AI is a bioinformatics platform that combines specialized AI agents with open-source tools to analyze multi-omics data through natural language.
When to use this Skill
Use Lobster when the user asks to:
Analyze single-cell RNA-seq data (QC, clustering, annotation, markers)
After asking user - Install a model: ollama pull gpt-oss:20b
Anthropic/Bedrock API error
Check API key validity in .env
NCBI API not configured
Add NCBI_API_KEY to .env (optional)
Workspace not writable
Check directory permissions
Quick status checks:
# Show configuration values (masked)
lobster config-show
# Show subscription tier and available agents
lobster status
Usage
Basic syntax
# Single query (non-interactive)
lobster query "<natural language request>"# With custom workspace
lobster query --workspace /path/to/workspace "<request>"# With reasoning mode (for complex tasks)
lobster query --reasoning "<request>"
Session continuity (multi-turn conversations)
Lobster supports conversation continuity via --session-id, enabling follow-up questions that reference previous context either by setting sessin-id to latest or a string of your choice:
# default session
lobster query "Search PubMed for CRISPR papers"# Output: Session: session_20241208_150000 (use --session-id latest for follow-ups)# then follow up with
lobster query --session-id latest "Download the first dataset from that search"#or use custom session id
lobster query --session-id "crispr_search_1""Search PubMed for CRISPR papers"#follow up with
lobster query --session-id "crispr_search_1""show me metadata from the first paper"
Best practices:
Always use --session-id latest for follow-up queries
Session files are saved in workspace as session_*.json
Use same --workspace for related queries to maintain context
Session contains conversation history, not tool execution state
Workspace-based sessions:
# Project 1: Cancer research
lobster query --workspace ~/cancer-project "Search for breast cancer datasets"
lobster query --workspace ~/cancer-project --session-id latest "Download the best one"# Project 2: Immunology (separate session)
lobster query --workspace ~/immuno-project "Search for T cell datasets"
lobster query --workspace ~/immuno-project --session-id latest "Analyze that"
Common patterns
Single-cell analysis:
lobster query "Download GSE109564 and perform quality control"
lobster query "Cluster the dataset and find marker genes"
lobster query "Create UMAP visualization colored by cell type"
Literature mining:
lobster query "Search PubMed for CRISPR screens in cancer"
lobster query "Find papers about CAR-T therapy and extract their GEO datasets"
lobster query "Get the full text and methods section for PMID:12345678"
lobster query "Load counts.csv and run differential expression analysis"
lobster query "Perform batch correction on the loaded dataset"
lobster query "Generate volcano plot for DE results"
Quality control:
lobster query "Assess quality metrics for the loaded dataset"
lobster query "Filter cells with <200 genes or >8000 genes"
lobster query "Identify doublets using scrublet"
Output handling
Lobster outputs are saved in the workspace directory (default: .lobster_workspace/):
Key files to check:
*.h5ad - Processed datasets (AnnData format)
*.html - Interactive visualizations
*.png - Static plots for publications
*.csv - Exported data tables
*.json - Metadata and provenance
To read results:
# List workspace filesls -lh .lobster_workspace/
# Read specific outputscat .lobster_workspace/analysis_summary.json
Integration workflow
Example 1: Analyze dataset and extract results
# Step 1: Run analysis
lobster query --session-id "gse109564""Download GSE109564, run QC, and cluster cells"# Step 2: Check outputsls .lobster_workspace/*.h5ad
ls .lobster_workspace/*.html
# Step 3: Extract specific data
lobster query --session-id "gse109564""Export cluster markers to CSV"# Step 4: Use results in your code# Results are now in .lobster_workspace/markers.csv
Example 2: Literature mining workflow
# Step 1: Find papers
lobster query "Search for papers about immune checkpoint inhibitors in melanoma"# Step 2: Extract datasets
lobster query "Extract all GEO dataset IDs from the cached papers"# Step 3: Validate datasets
lobster query "Check which datasets have cell_type and treatment metadata"# Step 4: Download best match
lobster query "Download the dataset with most samples"
Advanced features
Export reproducible notebooks:
lobster query "Export the analysis pipeline as a Jupyter notebook"# Creates a Papermill-compatible notebook in workspace
Workspace management:
# Use custom workspace per project
lobster query --workspace ./project1-data "Analyze counts.csv"
lobster query --workspace ./project2-data "Analyze other-counts.csv"
