Association extraction from biomedical literature based on representation and transfer learning.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Extracting biological relations from biomedical literature is crucial for personalized medicine based on genomic profiles.
• Current methods often rely on handcrafted biomedical features, limiting their generalizability.

Purpose of the Study:

• To develop a novel sentence-level attention-based deep neural network for predicting semantic relationships between medical entities.
• To evaluate the model's performance using a transfer learning paradigm without domain-specific features.

Main Methods:

• Utilized a deep neural network with a sentence-level attention mechanism.
• Employed transfer learning with pre-trained embedding models on PubMed and PMC papers.
• Focused solely on sentence information, transforming it into improved embedding vectors.

Main Results:

• Achieved competitive performance compared to state-of-the-art methods.
• Demonstrated the effectiveness of relying solely on sentence information.
• Eliminated the need for handcrafted biomedical features.

Conclusions:

• The proposed deep neural network effectively predicts biological relations from text.
• This approach offers a promising direction for advancing personalized medicine through automated literature analysis.
• The model's ability to perform without domain-specific features enhances its applicability and reduces development effort.