Microbial named entity recognition and normalisation for AI-assisted literature review and meta-analysis.

Deep learning models trained on a novel microbiome-specific corpus achieved 96% F1-score for NER and 91% accuracy for entity linking, significantly outperforming traditional pipelines.

• Why it matters: Accurate extraction and normalization of microbial entities from biomedical literature are essential for efficient meta-analyses and advancing microbiome research, addressing limitations of manual curation and general language models.
• What they did: Researchers created the first microbiome-specific text corpus with over 90,000 annotations, trained deep learning models including BioBERT, and evaluated their performance across 6,927 full-text articles in 14 domains.
• The result: The models can annotate entire documents in just 7 seconds, enabling rapid, accurate literature analysis and providing open-source tools and datasets to facilitate further research and automation in microbiome studies.