DTF

DTF predicts synergistic interactions between anticancer drugs by integrating tensor factorization and a deep neural network to classify drug-pair synergy and impute missing entries in a drug–cell-line tensor.

Key Features:

• Integration of tensor factorization and deep learning: The tensor factorization component extracts latent features from drug synergy data while a deep neural network implements a binary classifier to predict synergy status.
• Latent feature extraction: Tensor factorization uncovers latent representations from existing drug–cell–line data organized as a tensor.
• Predictive performance: DTF achieved a precision-recall area under the curve (PR AUC) of 0.58 compared to 0.24 for a tensor-only method.
• Comparison to other methods: DTF showed comparable classification performance to DeepSynergy across several metrics.
• Imputation and discovery: DTF predicts missing entries in a drug–cell-line tensor and identified novel synergistic combinations for ten cell lines across five cancer types.
• Literature validation: Some predicted synergies were corroborated by in vivo or in vitro studies reported in the literature.

Scientific Applications:

• Prioritization of drug combinations: In silico ranking of candidate anticancer drug pairs for follow-up experimental testing.
• Imputation of experimental data: Filling missing entries in drug–cell-line tensors to guide targeted experiments.
• Support for combination therapy discovery: Identifying synergistic pairs across multiple cancer types to inform combination therapy hypotheses.

Methodology:

DTF uses existing drug synergy data structured as a tensor, applies tensor factorization to extract latent features, and employs a deep neural network as a binary classifier to predict synergy and impute missing tensor entries.