All-paths graph kernel for protein-protein interaction extraction with evaluation of cross-corpus learning
2008

Graph Kernel for Protein-Protein Interaction Extraction

Sample size: 1000 publication Evidence: high

Author Information

Author(s): Airola Antti, Pyysalo Sampo, Björne Jari, Pahikkala Tapio, Ginter Filip, Salakoski Tapio

Primary Institution: Turku Centre for Computer Science (TUCS) and the Department of IT, University of Turku

Hypothesis

Can a graph kernel approach improve the extraction of protein-protein interactions from biomedical texts?

Conclusion

The graph kernel approach achieves state-of-the-art performance in protein-protein interaction extraction.

Supporting Evidence

  • The method achieved a 56.4 F-score and 84.8 AUC on the AImed corpus.
  • Cross-corpus evaluations showed that the method generalizes beyond the training data.
  • The study identifies several pitfalls in evaluating PPI extraction systems.

Takeaway

This study shows a new way to find connections between proteins in scientific texts using a special graph method, which works really well.

Methodology

The study uses a graph kernel approach to analyze dependency graphs of sentences containing protein names and evaluates the method on five publicly available PPI corpora.

Potential Biases

Differences in corpus characteristics may introduce bias in performance evaluation.

Limitations

The performance varies significantly across different corpora, making direct comparisons challenging.

Statistical Information

P-Value

p<0.05

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1186/1471-2105-9-S11-S2

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication