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

Abstract

Information about the interactions between chemical compounds and proteins is indispensable for understanding the regulation of biological processes and the development of therapeutic drugs. Manually extracting such information from biomedical literature is very time and resource consuming. In this study, we propose a computational method to automatically extract chemical–protein interactions (CPIs) from a given text. Our method extracts CPI pairs and CPI triplets from sentences, where a CPI pair consists of a chemical compound and a protein name, and a CPI triplet consists of a CPI pair along with an interaction word describing their relationship. We extracted a diverse set of features from sentences that were used to build multiple machine learning models. Our models contain both simple features, which can be directly computed from sentences, and more sophisticated features derived using sentence structure analysis techniques. For example, one set of features was extracted based on the shortest paths between the CPI pairs or among the CPI triplets in the dependency graphs obtained from sentence parsing. We designed a three-stage approach to predict the multiple categories of CPIs. Our method performed the best among systems that use non-deep learning methods and outperformed several deep-learning-based systems in the track 5 of the BioCreative VI challenge. The features we designed in this study are informative and can be applied to other machine learning methods including deep learning.

Extraction of chemical–protein interactions from the literature using neural networks and narrow instance representation

“…Differently from the works cited above, Lung et al [34] used traditional machine learning algorithms with handcrafted features, achieving an F1-score of 0.5671. As part of their approach, the authors manually built a dictionary with 1155 interaction words, which where mapped to the corresponding CPR type, to create CPI triplets.…”

“…Though, we did not follow this possibility leaving it as possible future work. Similar binary approaches were followed by Lung et al [33, 34] and Warikoo et al [44] who start by predicting if a CPR pair is positive.…”

“…They later improved this result to 0.637 with a revised pre-processing and by using more members in the ensemble, and equaled the best challenge F1-score (0.641) using a shift-reduce parser based network architecture [32]. Lung et al [33, 34] achieved an F1-score of 0.567 using traditional machine learning. Neural networks with attention mechanisms were also followed by Liu et al [35, 36] and Verga et al [37], but achieved lower results.…”

PPPred: Classifying Protein-phenotype Co-mentions Extracted from Biomedical Literature

“…e rules can be derived from manually annotated corpora using machine learning algorithms or de ned manually by a domain expert. Machine learning-based approaches are also employed for the relation extraction from biomedical text [17,20,22,38]. e machine learning category includes methods based on feature engineering, graph kernels, and deep learning.…”

“…We obtained these features based on (1) domain expertise and (2) informative features used with similar relation extraction problem [22]. e full list of engineered features and their value type (within parentheses) is as follows:…”