Identifying protein subcellular localisation in scientific literature using bidirectional deep recurrent neural...

Area of Science:

• Bioinformatics
• Computational Biology
• Plant Science

Background:

• The rapid growth of biological data necessitates advanced computational methods for extracting meaningful insights.
• Relationship Extraction is crucial for understanding semantic interactions between biological entities in scientific literature.

Purpose of the Study:

• To develop and evaluate deep neural network natural language processing (NLP) methods for predicting protein subcellular localization in plants.
• To create a framework for extracting protein functional features from unstructured text with high accuracy.

Main Methods:

• Employed two deep neural network NLP methods: continuous bag of words (CBOW) and bi-directional long short-term memory (bi-LSTM).
• Applied the system to 1700 Arabidopsis protein subcellular localization studies from the SUBA dataset.
• Combined text pre-processing with relevant sentence extraction for NLP analysis.

Main Results:

• Achieved high prediction accuracy for protein name, subcellular localization, and experimental methodology (95.1% precision, 82.8% recall, 89.3% accuracy, 88.4% F1 score) on the SUBA corpus.
• Validated the model's wide applicability using the CropPAL database for crop species, demonstrating comparable performance.
• Successfully extracted protein functional features from unstructured literature data.

Conclusions:

• The developed NLP framework accurately extracts protein subcellular localization information from scientific literature.
• This approach enhances data dissemination and unlocks the potential of big data text analytics for generating new biological hypotheses.
• The model demonstrates broad applicability across different plant species.