mOWL: Python library for machine learning with biomedical ontologies.

Area of Science:

• Bioinformatics
• Machine Learning
• Ontology Engineering

Background:

• Ontologies provide structured domain knowledge crucial for bioinformatics data annotation and integration.
• Machine learning methods increasingly leverage ontologies for background knowledge, particularly in complex biological applications.
• A gap exists in readily available software libraries for common primitives needed to effectively use ontologies in machine learning.

Purpose of the Study:

• To develop a Python library, mOWL, enabling machine learning applications with ontologies.
• To provide essential primitives for integrating ontological knowledge into machine learning workflows.
• To facilitate the development of novel ontology-based methods in biomedical research.

Main Methods:

• Developed mOWL, a Python library for machine learning with ontologies in the Web Ontology Language (OWL).
• Implemented ontology embedding methods to represent ontological information in vector spaces.
• Integrated methods for similarity computation, deductive inference, and zero-shot learning using these embeddings.

Main Results:

• mOWL successfully maps ontological information into vector spaces, preserving key properties and relations.
• Demonstrated mOWL's utility in knowledge-based prediction of protein-protein interactions using the Gene Ontology.
• Showcased mOWL's application in predicting gene-disease associations using phenotype ontologies.

Conclusions:

• mOWL provides a foundational software library for machine learning with ontologies in bioinformatics.
• The library supports advanced machine learning tasks, including inference and zero-shot learning, by leveraging ontological knowledge.
• mOWL is available as an open-source Python package, promoting wider adoption and development in the field.