CLEP

CLEP generates patient representations by integrating patient-level datasets with prior biological knowledge via knowledge graphs and embedding models to improve downstream analyses such as classification of patients versus healthy controls and interpretation of transcriptomics-derived disease features.

Key Features:

• Integration of Knowledge Graphs: Utilizes knowledge graphs that encode known biological interactions and relations across features relevant to the dataset.
• Patient Representation Generation: Introduces patients as unique nodes in the knowledge graph linked to their most characteristic features.
• Knowledge Graph Embedding Models: Employs embedding models to transform enriched knowledge graphs into patient representations that capture relationships between patient data and biological pathways or interactions.
• Versatility for Downstream Tasks: Produces representations applicable to downstream analyses including clustering and classification tasks.

Scientific Applications:

• Patient versus control classification: Enhances machine learning model performance for classifying patients from healthy controls compared with using raw transcriptomics data alone.
• Disease-associated feature interpretation: Facilitates interpretation of biological features characteristic of specific diseases or patient subgroups.
• Investigation of disease mechanisms: Supports analysis that advances understanding of disease mechanisms by contextualizing patient data within known biological interactions.

Methodology:

Map a patient-level dataset onto a pre-existing knowledge graph, add patients as nodes connected to their most relevant features, and apply knowledge graph embedding techniques to generate new patient representations that combine individual data with contextual biological information.