Self-Attention-Based Models for the Extraction of Molecular Interactions from Biological Texts

Area of Science:

• Bioinformatics
• Computational Biology
• Natural Language Processing

Background:

• Keeping up with scientific publications is challenging due to the increasing volume of research.
• Automated text mining is crucial for building large-scale molecular interaction maps and curating biological databases.
• Natural Language Processing (NLP) techniques, including Named Entity Recognition (NER) and Relationship Extraction (RE), are vital for mining biological literature.

Purpose of the Study:

• To review self-attention-based neural network models applied to biological literature mining.
• To explore models for molecular interaction extraction from biological texts published since 2019.
• To compare architectures and discuss limitations and opportunities in biological literature mining.

Main Methods:

• Literature review focusing on self-attention-based neural network architectures.
• Analysis of models operating at sentence or abstract levels for molecular interaction extraction.
• Comparative study of model architectures and discussion of current limitations.

Main Results:

• Self-attention models have significantly advanced NLP applications in biological text analysis.
• Recent models show promise in extracting molecular interactions at both sentence and abstract levels.
• Identified limitations highlight areas for future research in biological literature mining.

Conclusions:

• Self-attention neural networks offer powerful tools for automated molecular interaction extraction.
• Further research is needed to address limitations and enhance the accuracy of biological literature mining.
• These advancements support the development of comprehensive molecular interaction databases.