Improving Deep Learning Model for Drug Synergy Prediction via Topological Features

In cancer treatment, drug combination therapy has gained significant attention due to its potential for synergistic effects. However, traditional wet-lab screening methods are time-consuming and costly, limiting the scope of drug combination research. In recent years, computational biology approaches have been favored for their ability to efficiently predict drug synergies, but existing methods often overlook the complex network relationships between drug targets and cell line targets, which can affect model accuracy and generalization. To address this issue, we propose a method that combines deep learning algorithms with protein-protein interaction (PPI) network topological features, named DeepSynergyTF. This method improves the performance of deep learning models in predicting drug synergies by utilizing topological features. Specifically, by integrating PPI network datasets, we systematically analyze the distribution patterns and spatial proximity of drug targets and cell line targets, thereby constructing a series of improved network topological features. These features include key quantifiable metrics such as network betweenness centrality, closeness centrality, and network density. Experimental results show that our method significantly outperforms baseline methods, showing improvements across all performance metrics. This work highlights the potential of network topological features in predicting drug synergies.