A representation model for biological entities by fusing structured axioms with unstructured texts

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Structured semantic resources like ontologies formally define biological concepts and relationships.
• Current methods use plain text for representation learning (RL), hindering knowledge encoding due to differing formats of axioms and definitions.
• This limits the effectiveness of machine learning in biological knowledge discovery.

Purpose of the Study:

• To propose ERBK, a novel representation learning (RL) model for bio-entities.
• To improve the encoding of biological knowledge from structured semantic resources.
• To enhance machine learning applications in biology, particularly in zero-shot learning scenarios.

Main Methods:

• ERBK utilizes a knowledge graph embedding method to encode axioms.
• Deep convolutional neural models are employed to encode textual definitions.
• This approach differentiates the encoding of machine-readable axioms and human-understandable definitions.

Main Results:

• ERBK encodes more underlying biological knowledge compared to existing methods.
• The model demonstrates superior performance in predicting protein-protein interactions and gene-disease associations.
• ERBK maintains promising performance in zero-shot learning scenarios.

Conclusions:

• ERBK offers improved representations for biological knowledge.
• The method shows significant potential for advancing biological knowledge discovery and applications.
• The ERBK approach has generality and can be extended to other biological relation types.