find-collaborators

name	find-collaborators
description	Identify potential research collaborators based on topic overlap. Searches PubMed for researchers publishing on given topics, ranks them by publication output and topic relevance, and generates a ranked collaborator report with their research profiles.

Find Collaborators

Purpose

Discover and rank potential research collaborators based on specific research topics or keyword combinations. This skill systematically searches PubMed for active researchers in your research area, analyzes their publication profiles, identifies research overlap, and generates a ranked collaborator report with contact information and research summaries.

Use when / Not when

Use when

• You want to find collaborators in a specific research area (e.g., CRISPR, cancer genomics)
• You're expanding your research network within a specific field
• You need to identify co-authors for grant proposals (NIH, NSF, etc.)
• You're analyzing the competitive landscape for your research topic
• You want to identify industry partners or academic collaborators in your region/country
• You need to discover rising stars in your field (early-career researchers with high output)

Not when

• You need real-time trending topics (PubMed lag: 2-4 weeks)
• You require proprietary databases (MEDLINE-only publications included)
• You need researcher contact information (affiliations only; manual lookup required)
• Topics are highly specialized with <10 active researchers (search may return too few)
• You need patent analysis or non-publication research tracking
• Collaborators are outside life sciences (PubMed does not index computer science, engineering, etc.)

Expected inputs / outputs

Inputs

• Topic list (comma-separated): "CRISPR base editing,cancer genomics,personalized medicine"
• Command-line arguments: year range, result limits, species/country filters
• API calls: PubMed eSearch/eFetch (free, no key required)

Outputs

• collaborators.tsv: Tab-separated table with author name, affiliation, n_papers, topics_covered, top_journals, rank_score
• collaborator_profiles.md: Formatted markdown with one section per top-20 collaborator showing their full publication list and research summary
• topic_author_heatmap.png: Heatmap with authors on Y-axis, topics on X-axis, cell values = number of papers per author per topic
• ranking_chart.png: Horizontal bar chart of top-20 collaborators sorted by rank_score
• Summary to console: Total authors found, topic coverage stats, top 5 collaborators

Procedure

Step 1: Search for papers per topic

1. For each topic: construct PubMed query (default: topic as text search)
2. Call eSearch with year filters and max results per topic
3. Collect all PMIDs, deduplicate across topics
4. Track which topics each PMID maps to

Step 2: Fetch publication details

1. Batch eFetch PMID records (10 per request)
2. Parse XML for:
   • All author names and affiliations
   • Publication year, journal, title
   • Journal impact factor (if available from XML)
3. Extract author last name + initials for deduplication

Step 3: Build author-topic matrix

1. For each author: count papers per topic
2. Identify authors with presence in multiple topics
3. Filter by --min-papers threshold (default: 3 papers minimum)

Step 4: Rank collaborators

1. Compute collaboration score per author:

   score = (num_topics_covered * 10) + log1p(total_papers)
   

2. Apply country/institution filters if specified
3. Sort by score, return top N

Step 5: Augment author profiles (optional)

1. For top-30 authors: fetch their complete recent publication list
2. Extract research keywords (MeSH terms, abstracts)
3. Identify co-authors and collaboration frequency

Step 6: Generate outputs

1. Write TSV table of all qualifying authors
2. Create markdown report with top-20 profiles
3. Generate visualizations:
   • Author-topic heatmap
   • Ranking bar chart
   • (Optional) research keyword cloud per author

Key execution patterns

# Basic: Find collaborators in a single research area
python scripts/find_collaborators.py \
  --topics "CRISPR gene editing" \
  --years-back 5 \
  --outdir ./collaborators

# Multiple topics with overlap analysis
python scripts/find_collaborators.py \
  --topics "CRISPR base editing,epigenetics,precision medicine" \
  --years-back 3 \
  --max-per-topic 150 \
  --min-papers 5 \
  --outdir ./multi_topic_collab

# Find collaborators in a specific country
python scripts/find_collaborators.py \
  --topics "immunotherapy,cancer genomics" \
  --country-filter "Australia" \
  --years-back 2 \
  --outdir ./au_collab

# Find collaborators in human-focused research
python scripts/find_collaborators.py \
  --topics "CRISPR,genome editing" \
  --species-focus human \
  --exclude-institution "model organism" \
  --years-back 5 \
  --outdir ./human_crispr

# Large-scale collaborator discovery
python scripts/find_collaborators.py \
  --topics "machine learning,drug discovery,biotech" \
  --years-back 4 \
  --max-per-topic 300 \
  --min-papers 3 \
  --outdir ./ml_drug_discovery

Parameter decision guide

Parameter	Value	When to use	Rationale
--topics	Single broad topic	Initial scoping	Fast search, large result set
--topics	2-3 specific topics	Medium-specificity search	Good balance of precision and recall
--topics	4+ topics	Highly interdisciplinary work	More comprehensive coverage
--years-back	2	Recent trends only	Captures active researchers in hot topics
--years-back	3-5 (default)	Standard collaboration search	Captures established and emerging researchers
--years-back	10+	Established researcher tracking	Includes tenured researchers, may miss new entrants
--max-per-topic	50-100	Focused research areas	Fast search, top researchers by publication count
--max-per-topic	200+	Broad or popular topics	Comprehensive coverage, longer runtime
--min-papers	3 (default)	Standard threshold	Most researchers meet this; ~50-100 results typical
--min-papers	5-10	High-output researchers only	Focus on prolific authors; ~5-50 results typical
--species-focus	human	Clinical or translational work	Filters out pure model organism research
--species-focus	mouse	Basic research, model organism labs	Filters out human-centric research
--country-filter	Set (e.g., "USA")	Regional network building	Enables local collaboration
--exclude-institution	"text"	Remove noise	E.g., exclude "company name" or research consortia

Failure modes

Failure	Cause	Solution
Too many results (1000+)	Topic too broad or common	Add year filters, increase --min-papers, split topics
Too few results (<10)	Topic too specific or niche	Reduce --years-back, lower --min-papers to 1-2
Duplicate authors	Author name ambiguity (e.g., "Smith J")	Results include multiple people; manual review needed
Poor affiliation parsing	Garbled or non-standard affiliation strings	Extract institution manually from top results
Missing recent researchers	Lag in PubMed indexing (2-4 weeks)	Check preprints (bioRxiv) separately
API rate limit exceeded	Too many eFetch calls for large author set	Reduce --max-per-topic or increase delays
Empty heatmap	No author publishes in multiple topics	Topics are too specialized/non-overlapping
Heatmap too dense	Many authors × many topics	Filter to top-50 authors only
Missing keywords	MeSH terms incomplete or abstract not in XML	Script gracefully continues with available data