MDNNSyn: A Multi-Modal Deep Learning Framework for Drug Synergy Prediction

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in medicine

Background:

• Synergistic drug combination prediction is crucial in cancer therapy.
• Existing computational models often overlook complex biological relationships between drugs and cell lines.
• These relationships significantly influence drug synergy mechanisms.

Purpose of the Study:

• To propose a novel multi-modal deep learning framework, MDNNSyn, for predicting synergistic drug combinations.
• To integrate multi-source information and multi-modal features for enhanced prediction accuracy.
• To address the limitations of models focusing solely on pairwise drug-cell line interactions.

Main Methods:

• MDNNSyn utilizes a multi-modal deep learning approach.
• It extracts topology modality features using a multi-layer hypergraph neural network on drug synergy hypergraphs.
• Semantic modality features are constructed via a similarity strategy, and a gated neural network fuses these features for synergy score prediction.

Main Results:

• MDNNSyn achieved superior performance compared to five state-of-the-art methods on the DrugCombDB and Oncology-Screen datasets.
• The model obtained Area Under the Curve (AUC) scores of 0.8682 and 0.9013, respectively.
• Performance improvements were 3.70% and 2.71% over the second-best models, demonstrating enhanced predictive capability.

Conclusions:

• MDNNSyn effectively predicts potential synergistic drug combinations by integrating diverse biological data.
• The framework's ability to capture complex interactions enhances its utility in cancer drug discovery.
• Case studies confirm MDNNSyn's capability in identifying promising drug combinations for further investigation.