EM Connectomics Bibliometrics — Complete Documentation

Executive Summary

This analysis covers 7,503 papers on electron microscopy connectomics, extracted from OpenAlex (1977–2026). Core findings:

• 35,797 unique authors
• 85,589 citation relationships (directed network)
• 22 core journal club papers (using network importance metrics)
• 571 preprints (legitimate research, not duplicates)

Dataset Overview

Final Corpus: 7,503 Papers

Quality Assurance

• Duplicates removed: 420 (5.3% of original)
• Duplicate citations merged: 129,757
• ArXiv preprints merged into published: 1
• Final DOI duplicates: 0 ✓

Visualizations: What’s Included

1. PAPERS: Citation Network + Rankings

Where we draw the line:

• Includes: ALL 7,503 papers with metrics for each
• Network: 85,589 directed citation edges
• Layout: Force-directed D3 graph
• Ranked list: Sorted by composite importance score

Metrics per paper:

• In-degree: How many papers cite this (foundational signal)
• Out-degree: How many papers this cites (synthesis/scope signal)
• PageRank: Network importance via link structure (2,000 computed, derived for rest)
• HITS hub/authority: What type of influence
• Betweenness: Paper as bridge between communities
• K-core: Structural network density (0–32)
• Composite score: 80% PageRank + 20% k-core (weighted formula)

Visual encoding:

• Node size: Citation count
• Node color: Community membership
• Edge direction: Citation direction (A→B = A cites B)
• Hover: Full paper metadata

2. PEOPLE: Co-authorship Network + Rankings

Where we draw the line:

• Includes: ALL 35,797 authors appearing in corpus
• Network: 511,670 co-authorship edges (undirected)
• Ranked list: Top authors by metrics

Metrics per author:

• Number of papers: Productivity in connectomics
• Co-author count: Collaboration breadth
• PageRank (co-authorship): Influence in collaboration network
• Centrality: Position in author network
• Specialization: Connectomics focus vs. general neuroscience

Visual encoding:

• Node size: Number of papers
• Node color: Research specialization
• Edge weight: Number of collaborative papers
• Hover: Author details, top papers

3. JOURNAL CLUB: Core Selection

Where we draw the line:

• Base set: 7,503 papers (FULL CORPUS)
• Selection method: Role-weighted importance score
  • 80% composite score (PageRank + HITS + betweenness)
  • 20% k-core (structural position)
  • Boost: +0.1 for methods papers (out-degree > 50)
• Available thresholds:
  • Inclusive (≥0.10): 64 papers (emerging + everything)
  • Moderate (≥0.15): 33 papers (solid contributions)
  • Core (≥0.20): 22 papers (strong signal, recommended)
  • Strict (≥0.30): 15 papers (only very high importance)

Each paper includes:

• Tier badge (Platinum/Gold/Silver/Bronze)
• Citation count, year, DOI
• In-degree, out-degree, k-core
• Inferred role (foundational, methodological, synthesis, etc.)
• Community membership

Paper role interpretation:

• High in-degree / low out-degree: Landmark/foundational
  • Example: C. elegans connectome (631 cites, 0 refs within corpus)
• High out-degree / moderate in-degree: Methods/synthesis
  • Example: Graph analysis review (140 references)
• Balanced: Applied biology using methods on questions
• High out-degree (>50): Infrastructure (CV, ML, EM techniques)

4. K-CORE DECOMPOSITION

Where we draw the line:

• Includes: All 7,503 papers, k-values 0–32
• Color coding:
  • Red (k≥30): Inner core (highest structural density)
  • Orange (k=25–29): EM connectomics zone (emphasized)
  • Green (k=20–24): Bridge papers
  • Gray (k<20): Peripheral/emerging work

Interactive features:

• Filter by k-value range
• Node size: Citation count
• Hover: Paper details

Interpretation: K-core value shows how many neighbors each node has in the subgraph. Higher k = more structurally central in citation network.

5. FIELD EVOLUTION: Timeline

Where we draw the line:

• Period: 1977–2026 (50 years)
• Milestones: Key papers driving field forward
• Three phases:
  1. Theory (1977–2000): Foundational network science
  2. Methods (2000–2015): EM techniques, reconstruction tools
  3. Connectomes (2020+): Complete connectome datasets

Key Metrics Explained

In-Degree vs Out-Degree: Understanding Paper Roles

In-Degree (how many papers cite this):

• High in-degree = Foundational/landmark work
• Example: C. elegans connectome (631 citations)
• Stable importance, everyone cites it
• Signal: “This is a pillar the field is built on”

Out-Degree (how many papers this cites):

• High out-degree = Synthesis/methodology/infrastructure
• Example: Graph analysis review (140 references)
• Integrates field knowledge, enables others
• Signal: “This paper brings together the field”

Why both matter:

• Pure citation counts (in-degree only) bias toward foundational work
• Directedness reveals methods papers critical to infrastructure
• Methods papers might have moderate citations but extremely high out-degree
• Using both metrics: complete picture of importance

Composite Score: The Formula

Composite Score = 0.80 × (normalized PageRank/in-degree)
                + 0.20 × (normalized k-core value)

Why 80/20?

• PageRank captures network-structural importance (primary signal)
• K-core captures local clustering (secondary signal)
• Weighted toward PageRank because citation networks are robust
• K-core adds “structural density” perspective (how embedded in subgraph)

Data Characteristics

Paper Type Distribution (Final Corpus)

• Articles (journals): 5,229 (70%)
• Reviews: 1,523 (20%)
• Preprints: 571 (8%)
• Book chapters, books, others: 180 (2%)

Geographic/Organizational Representation

• Authors: 35,797 unique individuals
• Countries: ~150 (inferred from affiliations)
• Key institutions: Harvard, Cambridge, Janelia, MPI, etc.

Temporal Coverage

• Earliest: 1977 (foundational theory)
• Latest: 2026 (frontier papers)
• Peak activity: 2015–2025 (connectome era)

Known Limitations & Biases

1. OpenAlex coverage: May miss some pre-2005 papers, conference proceedings
2. Citation edge sampling: Limited to papers in our corpus (not global citations)
3. Author metrics: Only covers connectomics-related papers, not full author career
4. Clustering: Louvain community detection has some randomness
5. K-core: Requires connected subgraph, may not capture all structure
6. Recent papers: May be under-cited simply due to recency

Using This Analysis

For Literature Review

1. Start with Platinum tier papers (highest importance)
2. Read foundational papers (high in-degree, low out-degree)
3. Study synthesis reviews (high out-degree) for field overview
4. Dive into methodology papers (high out-degree, >50 refs)

For Understanding Field Structure

1. Look at communities in the network (groups of related work)
2. Follow citation paths (who cites whom)
3. Identify bridges (high betweenness papers connecting communities)

For Methods Papers

• Papers with out-degree > 50 are infrastructure
• These enable discovery but may have lower direct citations
• Critical for reproducibility and new technique adoption

Technical Details

Metric Computation

PageRank:

• Computed on full 7,503-node citation graph
• Damping factor: 0.85
• Converged values used for ranking

HITS (Hubs and Authorities):

• Hubs: Papers citing many important papers (high out-degree pattern)
• Authorities: Papers cited by many important papers (high in-degree pattern)
• Both computed on citation network

Betweenness Centrality:

• Shortest path frequency through each paper
• High value = bridge between communities

K-Core Decomposition:

• Maximal subgraph where each node has ≥k neighbors
• Repeated removal of degree-0 nodes until stable
• Value indicates minimum network connectivity requirement for inclusion

Communities (Louvain):

• Modularity-based clustering on citation network
• 30 communities detected
• 7 connectomics-relevant, 17 noise, 6 uncertain

Future Enhancements

Potential improvements:

• Extend author rankings beyond top 1,000
• Temporal dynamics (how importance changes over time)
• Cross-discipline comparison (how connectomics relates to broader neuroscience)
• Semantic clustering (topic modeling) alongside network analysis
• Interactive filtering by year, author, keyword
• Citation provenance tracking (why papers cite each other)

Citation

If you use this analysis, please cite:

Connectomics Bibliography Analysis. (2026). EM Connectomics Research Program. Data source: OpenAlex (https://openalex.org)

Questions & Issues

For questions about methodology or data quality, see:

• PIPELINE_COMPLETE.md (technical implementation)
• index.html (interactive visualizations)
• Each visualization’s inline documentation

Analysis Date: March 31, 2026
Corpus Version: corpus_final.json (7,503 papers, deduplicated)
Last Updated: March 31, 2026