Extracting chemical-protein interactions from literature using sentence structure analysis and feature engineering

Area of Science:

• Computational Biology
• Bioinformatics
• Natural Language Processing

Background:

• Understanding chemical-protein interactions (CPIs) is crucial for biological research and drug development.
• Manual extraction of CPIs from biomedical literature is labor-intensive and time-consuming.

Purpose of the Study:

• To develop and evaluate a computational method for the automatic extraction of CPIs from text.
• To extract both CPI pairs (chemical-protein) and CPI triplets (chemical-protein-interaction).

Main Methods:

• Utilized machine learning models trained on diverse sentence features, including simple and structure-based features.
• Employed sentence structure analysis, such as dependency graph shortest paths, for feature extraction.
• Designed a three-stage approach for predicting multiple categories of CPIs.

Main Results:

• The proposed method achieved superior performance compared to non-deep learning systems.
• Outperformed several deep-learning-based systems in the BioCreative VI challenge (track 5).
• Demonstrated the informativeness of designed features for various machine learning approaches.

Conclusions:

• The developed computational method offers an efficient and effective solution for automatic CPI extraction.
• The feature engineering techniques are broadly applicable to other machine learning and deep learning models.
• This work advances automated knowledge extraction in the biomedical domain.