The role of ontologies in machine learning: a case study of gene ontology

Abstract

Introduction. Ontologies as knowledgebases have been heavily applied in computational biological studies by implementing into ML models for purposes such as disease-gene associations identification.

Method. We conduct a case study using gene ontology (GO) annotation data and three ML models to replicate the prediction of autism spectrum disorder (ASD)-causing genes.

Analysis. Data were collected from GO and Simmons Foundation Autism Research Initiative (SFARI). The semantic similarities between GO annotation terms on gene products were calculated.

Results. The best-performing model can reach an AUC of .85, which means using GO annotation data for ASD disease-gene prediction can receive a significantly accurate result. However, we stress the importance of constructing knowledgebases in adapting to LLMs and the role of LIS professionals in curating community knowledge for interoperability and reuse.

Conclusions. Biomedical ontologies play a crucial role in the discovery of biomedical knowledge. Knowledge organization and computer science domains require more communication and synchronization in the face of emerging AI and ML technologies.