Denario Elite

A scientific computing skill for automated research workflows using Denario — the multiagent AI system designed to orchestrate scientific research from data analysis through publication, with agents for literature review, data processing, statistical analysis, figure generation, and manuscript drafting.

When to Use This Skill

Choose Denario Elite when:

• Automating multi-step scientific analysis workflows
• Orchestrating literature review, data processing, and visualization
• Building reproducible research pipelines with AI agent coordination
• Generating publication-ready figures and preliminary manuscript drafts

Consider alternatives when:

• You need a specific analysis tool (use the dedicated tool directly)
• You're doing manual, exploratory data analysis (use Jupyter notebooks)
• You need real-time collaboration on analysis (use shared notebooks)
• You need domain-specific pipelines (use Nextflow, Snakemake, etc.)

Quick Start

Copy
claude "Set up a Denario workflow to analyze my gene expression dataset"

Copy
from denario import ResearchWorkflow, Agent

# Define research workflow
workflow = ResearchWorkflow(
    title="Differential Gene Expression Analysis",
    data_sources=["counts_matrix.csv", "sample_metadata.csv"]
)

# Configure agents
workflow.add_agent(Agent.LITERATURE_REVIEW, {
    "query": "RNA-seq differential expression best practices 2024",
    "max_papers": 20
})

workflow.add_agent(Agent.DATA_PROCESSOR, {
    "pipeline": "deseq2",
    "normalization": "median_of_ratios",
    "filters": {"min_count": 10, "min_samples": 3}
})

workflow.add_agent(Agent.STATISTICIAN, {
    "test": "wald",
    "correction": "BH",
    "alpha": 0.05,
    "log2fc_threshold": 1.0
})

workflow.add_agent(Agent.VISUALIZER, {
    "plots": ["volcano", "heatmap", "pca", "ma_plot"],
    "format": "publication_ready"
})

# Execute workflow
results = workflow.run()
print(results.summary)

Core Concepts

Denario Agent Types

Agent	Role	Output
Literature Review	Search and summarize relevant papers	Literature summary
Data Processor	Clean, normalize, transform data	Processed datasets
Statistician	Run statistical tests and modeling	Test results, p-values
Visualizer	Generate figures and plots	Publication-ready figures
Writer	Draft manuscript sections	Text drafts
Reviewer	Check analysis validity	Quality report

Workflow Orchestration

Copy
# Agents communicate through shared context
workflow = ResearchWorkflow(title="My Study")

# Sequential pipeline
workflow.pipeline([
    ("data_ingestion", Agent.DATA_PROCESSOR),
    ("quality_control", Agent.REVIEWER),
    ("analysis", Agent.STATISTICIAN),
    ("visualization", Agent.VISUALIZER),
    ("manuscript", Agent.WRITER)
])

# Each agent receives output from the previous step
results = workflow.run()

# Access individual agent outputs
figures = results.get_agent_output("visualization")
manuscript = results.get_agent_output("manuscript")

Reproducibility Framework

Copy
# All workflow steps are logged and reproducible
workflow.export_provenance("provenance.json")

# Provenance includes:
# - Input data checksums
# - Parameter configurations
# - Software versions
# - Intermediate results
# - Execution timestamps

# Reproduce from provenance
reproduced = ResearchWorkflow.from_provenance("provenance.json")
reproduced.run()

Configuration

Parameter	Description	Default
agents	List of agents in the pipeline	Required
data_sources	Input data file paths	Required
output_dir	Directory for results	./results
figure_format	Output figure format	pdf
reproducibility	Record full provenance	true

Best Practices

1. Define clear analysis objectives upfront. Vague research questions produce vague results. Specify exactly what you want to test, which comparisons to make, and what significance thresholds to use before configuring the workflow.

2. Include a quality control agent. Always add a Reviewer agent that validates data quality, checks statistical assumptions, and flags potential issues. Automated analysis without quality checks can produce misleading results.

3. Export provenance for reproducibility. Every workflow should generate a complete provenance record — input data hashes, parameters, software versions, and intermediate results. This enables exact reproduction and peer review.

4. Validate agent outputs manually. AI-generated analysis can contain errors — verify statistical results against manual calculations for a subset of data, check that figures accurately represent the underlying data, and review any generated text for accuracy.

5. Version your workflows. Store workflow configurations in version control alongside your data analysis scripts. This tracks how your analysis evolved and enables colleagues to review or reproduce your work.

Common Issues

Agent produces unexpected output format. Verify the agent configuration matches your data type. A statistical agent configured for continuous data won't work correctly on categorical variables. Check the agent's expected input format against your actual data.

Workflow fails at intermediate step. Check agent logs for the specific error. Common causes: data format mismatch between agents, missing dependencies, or insufficient memory for large datasets. Fix the failing step and re-run from that point.

Generated figures don't meet journal requirements. Configure the Visualizer agent with specific journal requirements: DPI (300+), dimensions (column/page width), font sizes, and color accessibility. Most agents default to screen-quality output — publication requirements must be specified explicitly.