InfAcrOnt: calculating cross-ontology term similarities using information flow by a random walk

Area of Science:

• Biomedical Informatics
• Bioinformatics
• Computational Biology

Background:

• The number of biomedical ontologies like Gene Ontology (GO), Disease Ontology (DO), and Human Phenotype Ontology (HPO) has grown significantly.
• Calculating term similarities within ontologies is established, but cross-ontology similarities are less explored.
• Existing methods for cross-ontology similarity often rely solely on gene interactions and are primarily designed for GO, with unknown performance on other ontologies.

Purpose of the Study:

• To propose a novel method, InfAcrOnt, for inferring similarities between terms across different biomedical ontologies.
• To leverage the complete Gene Functional Interaction Network (GFIN) for enhanced cross-ontology similarity inference.
• To evaluate InfAcrOnt's performance on benchmark datasets and compare it with prior knowledge.

Main Methods:

• InfAcrOnt constructs a comprehensive term-gene-gene network integrating ontology annotations and GFIN.
• It models information flow within this network using a random walk approach to determine cross-ontology term similarities.
• The method was benchmarked on GO sub-ontologies using human and yeast datasets.

Main Results:

• InfAcrOnt achieved high performance with average AUC values of 0.9322 (human) and 0.9309 (yeast), and very low standard deviations.
• The method demonstrated superior performance compared to existing approaches on benchmark datasets.
• InfAcrOnt's results showed high correlations with prior knowledge for pairwise DO-HPO and DO-GO terms.

Conclusions:

• InfAcrOnt significantly enhances the accuracy of inferring similarities between terms across biomedical ontologies.
• The method effectively utilizes the GFIN, improving upon previous approaches.
• InfAcrOnt offers a robust solution for cross-ontology similarity analysis in the broader biomedical community.