figure-pipeline

Five-Step Figure Generation Pipeline

Generate publication-quality scientific figures through a structured pipeline with reference retrieval, planning, styling, generation, and self-critique. Inspired by PaperVizAgent's five-agent architecture.

When to Use

• Multi-panel composite figures (e.g., "Figure 1: A) boxplot, B) KM curve, C) heatmap, D) network")
• Figures that need to match a specific journal's style
• Complex visualizations where direct code generation often produces suboptimal layouts
• User explicitly asks for "publication-quality" or "journal-ready" figures
• Graphical abstracts

When NOT to use (overkill):

• Single simple plot → just write matplotlib/ggplot2 code directly
• Mechanism diagrams → use svg-scientific-figures skill
• Quick exploratory plots → direct code

---

Pipeline Steps

Step 1: RETRIEVE — Find Reference Figures

Search for published figures on the same topic to establish visual expectations:

bash: echo "=== PubMed figure captions ===" && \
curl -s "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&retmode=json&retmax=10&sort=relevance&term=TOPIC+AND+(figure+OR+visualization)" && \
echo -e "\n=== bioRxiv recent figures ===" && \
curl -s "https://api.biorxiv.org/details/biorxiv/2025-01-01/2026-03-18?jou=biorxiv" 2>/dev/null | head -c 2000

Also search for figure captions in full-text papers via Asta or Europe PMC:

curl -s "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=TOPIC+AND+FIG_TYPE:figure&format=json&pageSize=5"

From reference figures, extract:

• Figure types used (violin plot vs boxplot, clustered vs simple heatmap)
• Panel arrangement (2x2 grid, horizontal strip, vertical stack)
• Color schemes (which journal palette)
• Annotation patterns (significance brackets, gene labels, axis formatting)

Step 2: PLAN — Design Figure Composition

Create a structured figure plan:

{
  "figure_id": "Figure 1",
  "title": "THBS2 Expression and Survival Analysis Across Cancer Types",
  "layout": {
    "type": "grid",
    "rows": 2,
    "cols": 2,
    "width_cm": 17.5,
    "height_cm": 15
  },
  "panels": [
    {
      "id": "A",
      "type": "boxplot",
      "data": "TCGA pan-cancer expression",
      "x": "cancer_type",
      "y": "THBS2_expression_TPM",
      "notes": "Sort by median expression, highlight significant (red asterisks)"
    },
    {
      "id": "B",
      "type": "kaplan_meier",
      "data": "PAAD survival",
      "groups": "THBS2_high vs THBS2_low",
      "notes": "Include risk table, log-rank p, HR with 95%CI"
    },
    {
      "id": "C",
      "type": "heatmap",
      "data": "Immune cell correlation matrix",
      "notes": "Cluster by correlation, annotate r values for significant pairs"
    },
    {
      "id": "D",
      "type": "dot_plot",
      "data": "GO enrichment top 15 terms",
      "notes": "Color by p-value, size by gene count, order by enrichment score"
    }
  ],
  "shared_style": {
    "palette": "NPG",
    "font": "Arial",
    "label_size": 8,
    "title_size": 10
  }
}