A knowledge-driven conditional approach to extract pharmacogenomics specific drug-gene relationships from free text

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Pharmacogenomics (PGx) studies require accurate drug-gene relationship knowledge bases.
• Scientific literature contains valuable PGx information but is unstructured and difficult for machines to process.
• Automated extraction of structured PGx data from text is crucial for advancing drug response research.

Purpose of the Study:

• To develop an automated approach for extracting PGx-specific drug-gene relationships from scientific literature.
• To improve the machine understandability of pharmacogenomics knowledge.
• To create a foundation for more systematic and integrative computational PGx studies.

Main Methods:

• Developed a conditional relationship extraction approach using known drug-gene pairs as prior knowledge.
• Applied the method to extract PGx-specific drug-gene pairs from 20 million MEDLINE abstracts.
• Utilized a co-occurrence-based method as the underlying relationship extraction technique.

Main Results:

• The conditional approach significantly improved precision (0.345 vs. 0.11) and F1 measure (0.402 vs. 0.201) compared to an unconditioned method.
• Demonstrated the effectiveness of using prior knowledge to guide relationship extraction.
• Achieved a recall of 0.481 with the conditional approach.

Conclusions:

• The conditional relationship extraction method effectively extracts PGx-specific drug-gene relationships from biomedical literature.
• This approach enhances the creation of structured, machine-understandable knowledge bases for pharmacogenomics.
• The method is generalizable to extracting other types of relationships with existing knowledge bases.