Trawling the bibliome
Wynand Alkema
WC11 Maastricht 25-08-2021
For 3R related research
Protection of animals used for scientific purposes
– Legislation to protect experimental animals
since 1986 (harmonisation of internal market)
– Directive 2010/63/EU:
• Harmonise practices
• Implement 3Rs and improve animal
welfare
• Improve transparency and enforcement
Main findings from a survey:
No budget/time for specific 3R search
Existing relevant 3R information not found / not used
The problem
• How can we identify articles that describe methods
for Replacement, Refinement and Reduction?
– Often not explicitly described as such.
– Hard to define the correct search terms.
• As a consequence the search is
– Elaborate
– Time consuming
– Biased
– Incomplete
3R
Solution : Mine the bibliome
A Boolean search query
• Dependent on exact keywords
used.
• Even a very restrictive query yields
10s to 100s of papers.
• How to rank and prioritize?
• How to identify 3-R methods?
TenWise Knowledge map (KMAP)
> 0.5 million biological concepts Knowledge Map
Large scale text mining
Abstracts Full text
Computer assisted Knowledge map
generation
> 200 million biological relations
Connected to underlying literature
How are concepts connected?
Automated relationship mining
Filtering on 3R related words
https://kmine.tenwiseapps.nl/
The problem : Which keywords?
de Vries R and P Whaley. In Vitro Critical Appraisal Tool (IV-CAT): tool development
protocol (1.0.0). Zenodo 2018, DOI: 10.5281/zenodo.1493498.
Ordering and ranking the literature
Can we build in intelligent search engine for 3R
papers, based on PubMed?
Machine Learning Approach
Based on a set of known
positive and negative papers,
a computer model is trained
that can predict whether a
paper is about a 3R subject.
3865 papers describing
Alternative Animal Tests (Mesh
term)
A random abstract set
A computer generated concept map
Proof of Principle
List of top scoring hits with the model
Examine top scoring hits
Future steps
Alternative models
Conclusions & Acknowledgments
Automated, high-throughput
literature mining combined
with AI based classification can
help in identifying the 3R
needles in the literature
haystack.
Acknowledgements
• Merel Ristkes Hoitinga (SYRCLE, Netherlands)
• Rob de Vries (SYRCLE, Netherlands)
• Judith van Luijk (SYRCLE, Netherlands)
• Brun Ulfhake (Karolinska Institute, Sweden)
• Rafael Frias (Karolinska Institute, Sweden)
• Jennifer Stone (ANU, Australia)
• Brett Lidbury (ANU, Australia)
• Nils Hijlkema (TenWise)

Wc11 talk trawling_bibliome_3_r_alkema_25082021

  • 1.
    Trawling the bibliome WynandAlkema WC11 Maastricht 25-08-2021 For 3R related research
  • 2.
    Protection of animalsused for scientific purposes – Legislation to protect experimental animals since 1986 (harmonisation of internal market) – Directive 2010/63/EU: • Harmonise practices • Implement 3Rs and improve animal welfare • Improve transparency and enforcement
  • 3.
    Main findings froma survey: No budget/time for specific 3R search Existing relevant 3R information not found / not used The problem • How can we identify articles that describe methods for Replacement, Refinement and Reduction? – Often not explicitly described as such. – Hard to define the correct search terms. • As a consequence the search is – Elaborate – Time consuming – Biased – Incomplete 3R
  • 4.
    Solution : Minethe bibliome
  • 5.
    A Boolean searchquery • Dependent on exact keywords used. • Even a very restrictive query yields 10s to 100s of papers. • How to rank and prioritize? • How to identify 3-R methods?
  • 6.
    TenWise Knowledge map(KMAP) > 0.5 million biological concepts Knowledge Map Large scale text mining Abstracts Full text Computer assisted Knowledge map generation > 200 million biological relations Connected to underlying literature
  • 7.
    How are conceptsconnected?
  • 8.
  • 9.
    Filtering on 3Rrelated words https://kmine.tenwiseapps.nl/
  • 10.
    The problem :Which keywords? de Vries R and P Whaley. In Vitro Critical Appraisal Tool (IV-CAT): tool development protocol (1.0.0). Zenodo 2018, DOI: 10.5281/zenodo.1493498.
  • 11.
    Ordering and rankingthe literature Can we build in intelligent search engine for 3R papers, based on PubMed?
  • 12.
    Machine Learning Approach Basedon a set of known positive and negative papers, a computer model is trained that can predict whether a paper is about a 3R subject. 3865 papers describing Alternative Animal Tests (Mesh term) A random abstract set
  • 13.
  • 14.
  • 15.
    List of topscoring hits with the model
  • 16.
  • 17.
  • 18.
    Conclusions & Acknowledgments Automated,high-throughput literature mining combined with AI based classification can help in identifying the 3R needles in the literature haystack. Acknowledgements • Merel Ristkes Hoitinga (SYRCLE, Netherlands) • Rob de Vries (SYRCLE, Netherlands) • Judith van Luijk (SYRCLE, Netherlands) • Brun Ulfhake (Karolinska Institute, Sweden) • Rafael Frias (Karolinska Institute, Sweden) • Jennifer Stone (ANU, Australia) • Brett Lidbury (ANU, Australia) • Nils Hijlkema (TenWise)