FamPlex

FamPlex provides curated mappings of protein families and complexes to improve named entity recognition and grounding in biomedical text for extracting molecular mechanisms.

Key Features:

• Family and complex representation: Defines protein families and multi-subunit complexes as commonly encountered in biomedical text, including examples such as AKT and NF-κB.
• Manual curation: Entities are identified empirically from the literature and manually linked to uniform identifiers to support accurate grounding.
• Hierarchical organization: Supports multi-level membership definitions representing families, subfamilies, and gene-level constituents.
• Curated lexical patterns: Provides curated prefix and suffix patterns to improve named entity recognition and event extraction.
• Database cross-references: Links entities to equivalent entries in multiple related databases to enhance interoperability.
• Performance improvement: Evaluation with the REACH event extraction software showed grounding accuracy for families and complexes increased from 15% to 71%.
• System integration: Has been integrated into systems such as TRIPS/DRUM and tested on datasets including the BioCreative VI Bioentity Normalization Task.

Scientific Applications:

• Automated literature mining: Enables more accurate extraction of mechanistic information from biomedical literature by improving grounding of families and complexes.
• Data integration: Facilitates linking and integration of text-mined entities with curated interaction databases and other biological datasets.
• Support for molecular research: Improves reliability of downstream analyses in genomics, proteomics, and systems biology that depend on accurate entity grounding.

Methodology:

Manual curation from literature; multi-level hierarchical membership definitions; curated prefix and suffix patterns; mapping to equivalent entries in related databases; evaluation with REACH; integration into TRIPS/DRUM and testing on the BioCreative VI Bioentity Normalization Task.