HESML

HESML implements ontology-based semantic similarity measures and Information Content (IC) models to compute scalable semantic similarity across biomedical ontologies such as WordNet, SNOMED-CT, MeSH, OBO-based ontologies, and the Gene Ontology (GO).

Key Features:

• Efficient Ontology Representation: Provides an updated ontology representation optimized for efficiency and scalability compared to relational database representations and naive in-memory graph structures.
• Ancestors-based Shortest-Path Length (AncSPL): Implements AncSPL, a novel approximation of Dijkstra's algorithm for efficient shortest-path computation within taxonomies.
• Information Content (IC) Models: Enhances ontology-based Information Content (IC) models for semantic similarity calculations.
• Performance and Scalability: Outperforms existing libraries by several orders of magnitude on biomedical ontologies such as SNOMED-CT, MeSH, and GO, with AncSPL scaling linearly with the dimension of the common ancestor subgraph independent of overall ontology size.
• Real-Time Computation: Enables path-based measures that can be up to six orders of magnitude faster than exact implementations in large ontologies like SNOMED-CT and GO.
• Supported Ontologies: Supports WordNet, SNOMED-CT, MeSH, Gene Ontology (GO), and other OBO-based ontologies.

Scientific Applications:

• Biomedical Text Mining: Computes scalable semantic similarity measures to support concept disambiguation, entity linking, and relationship extraction in biomedical text mining.
• Genomics and Ontology-Based Analysis: Enables rapid semantic similarity computation for gene and phenotype annotation comparison and other genomics applications that rely on large ontological datasets.

Methodology:

Implements AncSPL to approximate Dijkstra's shortest-path computations within taxonomies, uses an updated ontology representation for efficiency, and validates performance through comprehensive benchmarks.