MOLI

MOLI integrates somatic mutation, copy number aberration, and gene expression data using deep neural networks to predict drug responses.

Key Features:

• Multi-omics late integration: Performs late integration of somatic mutation, copy number aberration, and gene expression modalities.
• Deep neural network architecture: Uses deep neural networks to learn predictive representations from omics data.
• Type-specific encoding sub-networks: Employs modality-specific encoding sub-networks to learn distinct features for each omics type.
• Feature concatenation: Concatenates learned modality-specific features into a unified representation for downstream prediction.
• Combined cost function: Optimizes the unified representation with a combined cost function comprising triplet loss and binary cross-entropy loss.
• Triplet loss for representation learning: Applies triplet loss to cluster responder sample representations together while separating them from non-responders.

Scientific Applications:

• Precision oncology: Enhances the clinical relevance of drug response predictions by integrating diverse omics data.
• Personalized treatment strategies: Contributes insights for individualized drug selection by improving prediction accuracy for responders versus non-responders.

Methodology:

Integrates somatic mutation, copy number aberration, and gene expression data via type-specific encoding sub-networks whose outputs are concatenated into a unified representation optimized using a combined triplet loss and binary cross-entropy loss.