pyMeSHSim

pyMeSHSim performs biomedical named entity recognition, normalizes UMLS concepts to MeSH (including MHs and SCRs), and computes semantic similarity between MeSH terms to support analysis of disease-associated phenotypes in genetic research.

Key Features:

• Named Entity Recognition: Employs MetaMap to identify biomedical named entities and produce UMLS concepts.
• Normalization and Mapping: Uses a proprietary curated dataset mapping UMLS concepts to MeSH main headings (MHs), supplementary concept records (SCRs), and their interrelations for term normalization.
• Semantic Similarity Algorithms: Implements four information content (IC)-based algorithms and one graph-based algorithm to compute semantic similarity between MeSH terms.
• Performance Evaluation: Evaluated on OMIM and GWAS phenotypes and reported recall > 0.94, precision > 0.56, and F1 > 0.70 compared to tools such as DNorm and TaggerOne.
• Enhanced Recognition with SCRs: Incorporates SCRs and a curation strategy for non-MeSH-synonymous UMLS concepts to improve recognition of OMIM phenotypes.
• Correlation with Existing Semantic Tools: Achieves high correlation (0.89–0.99) in semantic similarity calculations when compared to the meshes tool.

Scientific Applications:

• Genetic studies: Standardizes phenotype annotations for genes to facilitate genotype–phenotype analyses.
• Data integration: Normalizes biomedical entities to MeSH to improve interoperability across genomic and phenotype datasets.
• System genetics analysis: Quantifies semantic relatedness of phenotype terms to support systems-level interpretation of genetic associations.
• Phenotype-driven gene exploration: Enables comparison and aggregation of disease phenotypes associated with genes identified by OMIM, GWAS, or other methods.

Methodology:

Uses MetaMap for NER to produce UMLS concepts; maps UMLS concepts to MeSH MHs and SCRs via a proprietary curated dataset and interrelations; implements four IC-based and one graph-based semantic similarity algorithms; incorporates SCRs and a curation strategy for non-MeSH-synonymous UMLS concepts; evaluated on OMIM and GWAS phenotypes with comparisons to DNorm, TaggerOne, and meshes.