SUPREME

SUPREME integrates genomic, transcriptomic, proteomic, and other multiomics data using a Graph Convolutional Neural Network to generate patient-specific embeddings for cancer subtype prediction.

Key Features:

• Multiomics Data Integration: Integrates high-dimensional genomic, transcriptomic, proteomic, and other omics layers to capture complementary signals across data modalities.
• Graph Neural Network Utilization: Employs Graph Neural Networks / Graph Convolutional Neural Network to learn node embeddings from node features and graph-structured associations.
• Patient Embedding Generation: Generates patient-specific embeddings from multiple similarity networks derived from multiomics features and integrates these embeddings with raw features for classification.
• Performance Superiority: Demonstrated superior performance in breast cancer subtyping across three datasets, with inferred subtypes showing significant survival differences compared to labels based on a single data type.
• Model-Agnostic Application: Applied to two additional datasets and reported consistent outperformance of nine other approaches.

Scientific Applications:

• Cancer subtype prediction: Produces refined cancer subtype classifications by integrating multiomics data and GNN-derived embeddings.
• Patient stratification: Identifies patient groups with distinct biological profiles that correlate with survival differences.
• Ground truth refinement: Refines existing subtype labels by uncovering characteristics from integrated multiomics signals.

Methodology:

Constructs multiple patient similarity networks from multiomics features, generates patient-specific embeddings from those networks, and learns/integrates node embeddings with raw features using a Graph Convolutional Neural Network.