Related Experiment Video
Updated: Jun 10, 2026
05:47
Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
BioLMiner System: interaction normalization task and interaction pair task in the BioCreative II.5 challenge
Yifei Chen1, Feng Liu, Bernard Manderick
1Computational Modeling Lab, Vrije Universiteit Brussel, Brussels, Belgium. yifechen@vub.ac.be
Summary
The Biological Literature Miner (BioLMiner) system effectively extracts gene names, normalized gene names, and protein-protein interactions from scientific texts. Its machine learning-based subsystems achieved high performance in BioCreative II.5 challenge tasks.
Area of Science:
- Bioinformatics
- Computational Biology
- Natural Language Processing
Background:
- Biological literature contains vast amounts of information on genes, proteins, and their interactions.
- Efficiently extracting this information is crucial for advancing biological research.
- Existing text mining tools often require specialized adaptations for specific biological tasks.
Purpose of the Study:
- To develop and implement the Biological Literature Miner (BioLMiner) system for extracting key biological entities and relationships.
- To evaluate the performance of BioLMiner's subsystems on standardized biological literature mining tasks.
- To demonstrate the system's capability in identifying gene/protein names, normalized names, and protein-protein interactions.
Main Methods:
- BioLMiner employs a pipeline architecture with three subsystems: Gene Mention Recognizer (GMRer), Gene Normalizer (GNer), and Protein-Protein Interaction Pair Extractor (PPIEor).
- Subsystems are developed using machine learning techniques such as Support Vector Machines (SVMs) and Conditional Random Fields (CRFs) with informative features.
- The system integrates biological-specific resources and existing Natural Language Processing (NLP) tools.
Main Results:
- BioLMiner's GNer and PPIEor subsystems achieved high performance in the BioCreative II.5 challenge's Interaction Normalization Task (INT) and Interaction Pair Task (IPT), respectively.
- The Gene Mention Recognizer (GMRer) provided essential support for the performance of the INT and IPT subsystems.
- The developed methods in GNer and PPIEor demonstrated effective extensibility to complex BioCreative II.5 tasks.
Conclusions:
- The BioLMiner system is a robust and high-performing text mining solution for biological literature.
- The machine learning-based approach effectively extracts and normalizes biological entities and interactions.
- BioLMiner's architecture and methods show promise for advancing automated information extraction in biology.
