DeepSynergy

DeepSynergy predicts synergistic interactions between anti-cancer drugs using deep learning to prioritize effective combination therapies in oncology.

Key Features:

• Deep Learning Framework: Uses deep neural network models to analyze large-scale combination screening data.
• Input Data Utilization: Integrates chemical and genomic information and applies a normalization strategy to manage heterogeneity in the input data.
• Model Architecture: Employs conical layers within the neural network architecture to capture and model drug synergy effects.

Scientific Applications:

• Identification of synergistic drug combinations: Predicts drug pairs likely to exhibit synergy to guide selection of candidate combinations for further study.
• Prioritization for experimental validation: Ranks predicted synergistic interactions to focus experimental resources on the most promising combinations.

Methodology:

Integrates chemical and genomic inputs with normalization and a deep neural network architecture that includes conical layers; evaluated against Gradient Boosting Machines, Random Forests, Support Vector Machines, and Elastic Nets using the largest publicly available synergy dataset; achieved a 7.2% improvement in mean squared error over the second-best method, a mean Pearson correlation coefficient of 0.73 between predicted and measured values for novel combinations within explored drugs and cell lines, and a classification AUC of 0.90; all methods, including DeepSynergy, showed limited predictive power when extrapolating to unexplored drugs or cell lines due to constraints in dataset size and diversity.