A Gated Recurrent Unit based architecture for recognizing ontology concepts from biological literature

Area of Science:

• Biomedical Informatics
• Computational Biology
• Bioinformatics

Background:

• Ontology concept annotation of scientific literature is crucial for biological knowledge discovery.
• Applications include evolutionary phenotypes, rare human diseases, and protein function studies.
• Computational methods have evolved from lexical approaches to deep learning.

Purpose of the Study:

• To develop and evaluate state-of-the-art deep learning architectures for ontology concept annotation in scientific text.
• To improve the accuracy of automated literature annotation using Gated Recurrent Units.

Main Methods:

• Utilized deep learning architectures, specifically Gated Recurrent Units (GRUs).
• Trained and tested models on the Colorado Richly Annotated Full Text Corpus (CRAFT).
• Integrated external knowledge sources like NCBI's BioThesauraus and Unified Medical Language System (UMLS) to augment training data.

Main Results:

• Achieved a state-of-the-art performance with an F1 score of 0.84.
• Demonstrated the effectiveness of GRU-based models for ontology concept recognition.
• Showcased improved prediction accuracy through the augmentation of training data with external biological information.

Conclusions:

• Deep learning architectures significantly enhance the automatic recognition of ontology concepts from scientific literature.
• Augmenting models with external biological data beyond the gold standard corpus leads to distinct improvements in prediction accuracy.
• This approach facilitates large-scale comparative analyses and knowledge discovery in various biological domains.