Combining automatic table classification and relationship extraction in extracting anticancer drug-side effect pairs...

Area of Science:

• Biomedical Informatics
• Pharmacology
• Computational Biology

Background:

• Anticancer drug-associated side effect knowledge is fragmented across diverse data sources.
• A unified drug-side effect (drug-SE) knowledge base is crucial for advancing drug discovery, toxicity prediction, and repositioning.

Purpose of the Study:

• To develop and validate a two-step computational approach for extracting drug-SE pairs from high-profile oncological literature.
• To compare extracted knowledge with existing databases and analyze relationships between drug side effects and biological targets/indications.

Main Methods:

• Utilized a dataset of 31,255 tables from the Journal of Oncology (JCO).
• Employed a statistical classifier for table classification (SE-related vs. unrelated) followed by relationship extraction.
• Compared extracted drug-SE pairs with those from FDA drug labels and analyzed correlations with gene targets, metabolism, and disease indications.

Main Results:

• The table classifier achieved high performance (precision: 0.711, recall: 0.941, F1: 0.810).
• Extracted 26,918 drug-SE pairs from SE-related tables with a precision of 0.605 and recall of 0.460.
• 84.7% of JCO-derived drug-SE pairs were novel compared to FDA drug label databases.
• Found positive correlations between anticancer drug side effects and drug targets, metabolism genes, and disease indications.

Conclusions:

• The developed method effectively extracts valuable drug-SE information from oncological literature.
• The extracted knowledge base significantly complements existing resources like FDA drug labels.
• Anticancer drug side effects are systematically linked to drug targets, metabolism, and disease contexts, offering insights for precision medicine.