← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / scientific-agent-skills / skills / literature-review / references / database_strategies.md

Literature Database Search Strategies

This document provides comprehensive guidance for searching multiple literature databases systematically and effectively.

Available Databases and Skills

Biomedical & Life Sciences

PubMed / PubMed Central

bioRxiv / medRxiv

General Scientific Literature

arXiv

Semantic Scholar

Google Scholar

Specialized Databases

ChEMBL / PubChem

UniProt

KEGG (Kyoto Encyclopedia of Genes and Genomes)

COSMIC (Catalogue of Somatic Mutations in Cancer)

AlphaFold Database

PDB (Protein Data Bank)

Citation & Reference Management

OpenAlex

Dimensions

Search Strategy Framework

1. Define Research Question (PICO Framework)

For clinical/biomedical reviews: - Population: Who is the study about? - Intervention: What is being tested? - Comparison: What is it compared to? - Outcome: What are the results?

Example: "What is the efficacy of CRISPR-Cas9 gene therapy (I) for treating sickle cell disease (P) compared to standard care (C) in improving patient outcomes (O)?"

2. Develop Search Terms

Primary Concepts

Identify 2-4 main concepts from your research question.

Example: - Concept 1: CRISPR, Cas9, gene editing - Concept 2: sickle cell disease, SCD, hemoglobin disorders - Concept 3: gene therapy, therapeutic editing

List alternative terms, abbreviations, and related concepts.

Tool: Use MeSH (Medical Subject Headings) browser for standardized terms

Boolean Operators

Example: (CRISPR OR Cas9 OR "gene editing") AND ("sickle cell" OR SCD) AND therapy

Wildcards & Truncation

Example: genom* matches genomic, genomics, genome

3. Set Inclusion/Exclusion Criteria

Inclusion Criteria

Exclusion Criteria

4. Database Selection Strategy

Multi-Database Approach

Search at least 3 complementary databases:

  1. Primary database: PubMed (biomedical) or arXiv (computational)
  2. Preprint server: bioRxiv/medRxiv or arXiv
  3. Comprehensive database: Semantic Scholar or Google Scholar
  4. Specialized database: ChEMBL, UniProt, or field-specific

Database-Specific Syntax

Database Field Tags Example
PubMed [Title], [Author], [MeSH] "CRISPR"[Title] AND 2020:2024[DP]
arXiv ti:, au:, cat: ti:"machine learning" AND cat:q-bio.QM
Semantic Scholar title:, author:, year: title:"deep learning" year:2020-2024

Search Execution Workflow

  1. Run initial search with broad terms
  2. Review first 50 results for relevance
  3. Note common keywords and MeSH terms
  4. Refine search strategy
  1. Execute refined searches across all selected databases
  2. Export results in standard format (RIS, BibTeX, JSON)
  3. Document search strings and date for each database
  4. Record number of results per database

Phase 3: Deduplication

  1. Import all results into a single file
  2. Use search_databases.py --deduplicate to remove duplicates
  3. Identify duplicates by DOI (primary) or title (fallback)
  4. Keep the version with most complete metadata

Phase 4: Screening

  1. Title screening: Review titles, exclude obviously irrelevant
  2. Abstract screening: Read abstracts, apply inclusion/exclusion criteria
  3. Full-text screening: Obtain and review full texts
  4. Document reasons for exclusion at each stage

Phase 5: Quality Assessment

  1. Assess study quality using appropriate tools:
  2. RCTs: Cochrane Risk of Bias tool
  3. Observational: Newcastle-Ottawa Scale
  4. Systematic reviews: AMSTAR 2
  5. Grade quality of evidence (high, moderate, low, very low)
  6. Consider excluding very low-quality studies

Search Documentation Template

Required Documentation

All searches must be documented for reproducibility:

## Search Strategy

### Database: PubMed
- **Date searched**: 2024-10-25
- **Date range**: 2015-01-01 to 2024-10-25
- **Search string**:
  ```
  ("CRISPR"[Title] OR "Cas9"[Title] OR "gene editing"[Title/Abstract])
  AND ("sickle cell disease"[MeSH] OR "SCD"[Title/Abstract])
  AND ("gene therapy"[MeSH] OR "therapeutic editing"[Title/Abstract])
  AND 2015:2024[Publication Date]
  AND English[Language]
  ```
- **Results**: 247 articles
- **After deduplication**: 189 articles

### Database: bioRxiv
- **Date searched**: 2024-10-25
- **Date range**: 2015-01-01 to 2024-10-25
- **Search string**: "CRISPR" AND "sickle cell" (in title/abstract)
- **Results**: 34 preprints
- **After deduplication**: 28 preprints

### Total Unique Articles
- **Combined results**: 217 unique articles
- **After title screening**: 156 articles
- **After abstract screening**: 89 articles
- **After full-text screening**: 52 articles included in review

Advanced Search Techniques

Prioritizing High-Impact Papers (CRITICAL)

Always prioritize papers based on citation count, venue quality, and author reputation. Quality matters more than quantity.

Citation Metrics in Database Searches

Use citation counts to identify influential work:

Paper Age Citations Classification
0-3 years 20+ Noteworthy
0-3 years 100+ Highly Influential
3-7 years 100+ Significant
3-7 years 500+ Landmark
7+ years 500+ Seminal
7+ years 1000+ Foundational

Database-Specific Citation Features: - Google Scholar: Sort by citation count, use "Cited by" feature - Semantic Scholar: "Highly Influential Citations" metric, citation velocity - OpenAlex: Citation counts, citation context analysis - PubMed: Use "Cited by" in PMC, check citation counts via Google Scholar

Filtering by Journal Quality

Prioritize papers from higher-tier venues:

Tier 1 (Always Prefer): - Nature, Science, Cell, NEJM, Lancet, JAMA, PNAS - Nature Medicine, Nature Biotechnology, Nature Methods - Search tip: source:Nature or journal:Nature in Google Scholar

Tier 2 (High Priority): - High-impact specialized journals (Impact Factor >10) - Top conferences: NeurIPS, ICML, ICLR, CVPR, ACL

Tier 3 (Include When Relevant): - Respected field-specific journals (IF 5-10)

PubMed Journal Filtering:

"Nature"[Journal] OR "Science"[Journal] OR "Cell"[Journal]

Google Scholar Journal Filtering:

source:Nature source:Science source:Cell

Leveraging "Cited by" Features

Finding Influential Work: 1. Start with a known key paper 2. Click "Cited by" to find papers that cite it 3. Sort citing papers by their citation count 4. Highly-cited citing papers indicate important follow-up work

Identifying Seminal Papers: 1. Search your topic broadly 2. Note which papers appear repeatedly in reference lists 3. Papers cited by many of your results are likely seminal 4. Check citation counts to confirm influence

Semantic Scholar Features: - "Highly Influential Citations" shows citations that significantly built on the paper - "Citation Velocity" shows recent citation growth - Paper recommendations based on citation networks

Citation Chaining

Find papers that cite a key paper: - Use Google Scholar "Cited by" feature - Use OpenAlex or Semantic Scholar APIs - Identifies newer research building on seminal work - Tip: Sort by citation count to find the most influential follow-up work

Review references in key papers: - Extract references from included papers - Search for highly cited references (500+ citations for older papers) - Identifies foundational research - Tip: Focus on references that appear in multiple papers' bibliographies

Snowball Sampling

  1. Start with 3-5 highly relevant papers from Tier-1 venues
  2. Extract all their references
  3. Check which references are cited by multiple papers
  4. Review those high-overlap references - these are likely seminal
  5. Repeat for newly identified key papers
  6. Prioritize papers with high citation counts at each step

Follow prolific and reputable authors in the field: - Search by author name across databases - Check author profiles (ORCID, Google Scholar) for h-index and publication venues - Review recent publications and preprints - Prefer authors with multiple Tier-1 publications and high h-index (>40) - Look for senior authors who are recognized field leaders

Related Article Features

Many databases suggest related articles: - PubMed "Similar articles" - Semantic Scholar "Recommended papers" - Use to discover papers missed by keyword search - Filter recommendations by citation count and venue quality

Quality Control Checklist

Before Searching

During Searching

After Searching

Common Pitfalls to Avoid

  1. Too narrow search: Missing relevant papers

  2. Solution: Include synonyms, related terms, broader concepts

  3. Too broad search: Thousands of irrelevant results

  4. Solution: Add specific concepts with AND, use field tags

  5. Single database: Incomplete coverage

  6. Solution: Search minimum 3 complementary databases

  7. Ignoring preprints: Missing latest findings

  8. Solution: Include bioRxiv, medRxiv, or arXiv

  9. No documentation: Irreproducible search

  10. Solution: Document every search string, date, and result count

  11. Manual deduplication: Time-consuming and error-prone

  12. Solution: Use search_databases.py script

  13. Unverified citations: Broken DOIs, incorrect metadata

  14. Solution: Run verify_citations.py on final reference list

  15. Publication bias: Only including published positive results

  16. Solution: Search trial registries, contact authors for unpublished data

Example Multi-Database Search Workflow

# Example workflow using available skills

# 1. Search PubMed via gget
search_term = "CRISPR AND sickle cell disease"
# Use gget search pubmed search_term

# 2. Search bioRxiv
# Use gget search biorxiv search_term

# 3. Search arXiv for computational papers
# Search arXiv with: cat:q-bio AND "CRISPR" AND "sickle cell"

# 4. Search Semantic Scholar via API
# Use semantic scholar API with search query

# 5. Aggregate and deduplicate results
# python search_databases.py combined_results.json --deduplicate --format markdown --output review_papers.md

# 6. Verify all citations
# python verify_citations.py review_papers.md

# 7. Generate final PDF
# python generate_pdf.py review_papers.md --citation-style nature

Resources

MeSH Browser

https://meshb.nlm.nih.gov/search

Boolean Search Tutorial

https://www.ncbi.nlm.nih.gov/books/NBK3827/

Citation Style Guides

See references/citation_styles.md in this skill

PRISMA Guidelines

Preferred Reporting Items for Systematic Reviews and Meta-Analyses: http://www.prisma-statement.org/