Chemical-induced disease relation extraction with various linguistic features

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Identifying chemical-disease relationships is vital for drug discovery and therapy development.
• Manual literature mining for these relations is time-consuming and difficult to maintain.
• The BioCreative-V challenge focused on automatic extraction of chemical-induced disease (CID) relations.

Purpose of the Study:

• To develop an automated system for extracting chemical-induced disease (CID) relations.
• To improve biocuration efficiency and support drug discovery initiatives.

Main Methods:

• A machine learning system using maximum entropy models and linguistic features was developed.
• Hypernym relations from Medical Subject Headings (MeSH) were incorporated for enhanced accuracy.
• Relation extraction was performed at both mention and document levels, with results merged for final output.

Main Results:

• The system achieved an F-score of 60.4% on the development dataset.
• An F-score of 58.3% was obtained on the test dataset using gold-standard entity annotations.
• The approach demonstrated effective extraction of CID relations.

Conclusions:

• The developed machine learning system offers a promising solution for automated CID relation extraction.
• Leveraging linguistic features and MeSH hypernyms improves extraction performance.
• This work contributes to efficient biocuration and advances in drug discovery research.