Visual Diagnostics

Generated: December 5, 2025

These figures follow Tufte’s principles for the visual display of quantitative information: maximizing data-ink ratio, clear labeling, and letting the data speak.

1. Treatment Cascade Over Time

How knowledge spillovers spread through the gene-disease network over time. Direct GWAS discoveries seed the network, then cascade outward through 1-hop, 2-hop, and 3-hop connections.

Treatment Cascade

Key insight: Treatment coverage accelerates after 2010, with most spillover pairs treated through 2-hop and 3-hop connections.

2. Event Study: Patents Around Treatment

Small multiples showing patent activity in the years before and after treatment, separately for each spillover level.

Event Study

Key insight: Look for pre-trends (should be flat before t=0) and treatment effects (changes after t=0).

3. Probability Threshold Analysis

The 99% BioBERT probability threshold is a key data quality decision. These panels show the trade-off between precision and coverage.

Probability Threshold

Panel A: Distribution of probabilities (most are high confidence) Panel B: Data retention curve showing steep drop-off above 95% Panel C: Unique patent coverage at different thresholds

4. Treatment Coverage Breakdown

The composition of the 148 million observation panel by treatment status.

Treatment Coverage

Key insight: 90% of observations are untreated (control group), 10% are treated across different spillover levels.

Aggregate patent activity in the panel across years.

Patent Trends

Key insight: Patent volume shows a quadratic growth pattern, accelerating in recent years.

6. Spillover Network Schematic

Conceptual visualization of the spillover structure. Knowledge flows outward from direct GWAS discoveries through the gene-disease network.

Spillover Schematic

Structure:

  • Direct GWAS: 35K pairs (core)
  • 1-hop: 189K pairs (immediate neighbors)
  • 2-hop: 814K pairs (second-degree connections)
  • 3-hop: 1.02M pairs (third-degree connections)

7. Concentration Analysis (Lorenz Curves)

Are patents concentrated among a few genes/diseases, or spread evenly?

Concentration

Key insight: Patent activity is highly concentrated. The top 10% of genes and diseases account for a disproportionate share of all patents.

8. Treatment Timing Distribution

When do gene-disease pairs first become treated at each spillover level?

Treatment Timing

Key insight: Direct GWAS discoveries are relatively steady over time, while spillover treatments show more variation year-to-year.

9. Summary Dashboard

All key metrics in one view.

Dashboard

Data Summary

Metric Value
Panel rows 148,124,655
Gene-disease pairs 7,053,555
Unique genes 14,368
Unique diseases 2,435
Unique patents 92,611
Years 2000-2020 (21)
Treated pairs 2,062,294 (29%)
Treated observations 14,849,919 (10%)

See also: Detailed Results