CLEP: a hybrid data- and knowledge-driven framework for generating patient representations.

Area of Science:

• Biomedical informatics
• Machine learning applications in healthcare
• Computational biology

Background:

• Machine learning and AI are increasingly used in biomedicine, but their effectiveness relies heavily on training data quality.
• Biomedical data's complexity and high dimensionality necessitate methods combining prior biological knowledge with patient-specific information.

Purpose of the Study:

• To introduce CLinical Embedding of Patients (CLEP), a novel approach for generating enhanced patient representations.
• To leverage both prior biological knowledge and patient-level data for improved machine learning utility in healthcare.

Main Methods:

• CLEP integrates patient data into a knowledge graph by adding patients as nodes linked to their features.
• Knowledge graph embedding models are then used to create new patient representations.
• The approach is released as an open-source Python package.

Main Results:

• CLEP significantly improved the performance of various machine learning models in classifying patients versus healthy controls compared to using raw transcriptomics data.
• Incorporating patients into the knowledge graph facilitated the interpretation and identification of disease-specific biological features.

Conclusions:

• CLEP offers a powerful method for generating informative patient representations by combining prior knowledge and patient data.
• This approach enhances downstream machine learning tasks and aids in biological interpretation for biomedical research.