BioNER

BioNER performs biomedical named entity recognition to extract and identify entities such as genes, proteins, and diseases from biomedical literature using neural network-based multi-task learning with a cross-sharing structure.

Key Features:

• Multi-Task Learning: Employs multi-task learning to learn and utilize features from multiple datasets simultaneously, improving generalization across entity types.
• Cross-Sharing Structure: Implements a cross-sharing structure to integrate and share features between tasks during training.
• Dataset Pair Optimization: Includes mechanisms to evaluate dataset pairs to identify optimal combinations that improve multi-task learning.
• Robustness to Dataset Size Variations: Maintains positive performance under reduced dataset sizes, demonstrating robustness to limited training data.
• Detailed Analysis and Guidance: Provides analyses of inter-entity influences and recommended dataset pairings to guide multi-task training decisions.

Scientific Applications:

• Biomedical literature mining: Extraction of genes, proteins, and diseases from biomedical literature for large-scale literature mining.
• Information extraction: Supports downstream information extraction tasks by improving entity recognition accuracy.
• Knowledge discovery: Facilitates knowledge discovery by providing more accurate entity-level signals for downstream analyses.
• Data integration: Supports data integration in bioinformatics by standardizing entity identification across datasets.
• Multi-entity recognition: Handles multiple entity types simultaneously to assist analyses of complex biomedical datasets.

Methodology:

Trains neural network models using a multi-task learning framework that incorporates a cross-sharing structure and includes dataset-pair evaluation mechanisms; implementation details specify Python 3+, PyTorch (versions < 1.0), and Gensim (version 12.0 or higher).