SWnet

SWnet predicts drug responses by integrating gene expression, genetic mutations, and chemical-structure data with deep learning to support precision cancer medicine.

Key Features:

• Integration of Multi-Omics Data: Combines gene expression profiles, genetic mutations, and chemical structures and extracts genomic features using cancer-related genes and L1000 landmark genes.
• Multi-task Convolutional Architecture: Employs a multi-task convolutional architecture that integrates an extended Graph Neural Network (GNN) for molecular graphs and a Convolutional Neural Network (CNN) for gene expression and mutation data.
• Cheminformatics Features: Incorporates cheminformatics-derived features from PubChem and ChEMBL to represent compound properties.
• Advanced Learning Mechanisms: Uses multi-task learning and self-attention mechanisms to model similarities between compounds and improve predictive performance.

Scientific Applications:

• Drug Sensitivity Prediction: Predicts drug sensitivity on datasets such as the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) and reports superior performance versus existing methods.
• Precision Medicine: Supports precision oncology by enabling individualized drug-response predictions for treatment selection based on comprehensive genomic and chemical data.

Methodology:

Extracts genomic features from cancer-related genes and L1000 landmark genes, represents compounds as molecular graphs and cheminformatics features from PubChem/ChEMBL, processes molecular graphs with an extended GNN and genomic data with a CNN within a multi-task convolutional architecture, and applies multi-task learning with self-attention to model compound similarities.