PRODeepSyn: predicting anticancer synergistic drug combinations by embedding cell lines with protein-protein...

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Drug combinations offer a promising cancer treatment strategy over monotherapy.
• Discovering synergistic drug combinations is hindered by combinatorial complexity.
• Deep learning shows potential for predicting drug synergy, often using omics data.

Purpose of the Study:

• To develop a novel deep learning method, PRODeepSyn, for predicting anticancer synergistic drug combinations.
• To integrate biological network data with omics data for improved cell line feature representation.
• To enhance the prediction accuracy of synergistic drug combinations.

Main Methods:

• PRODeepSyn utilizes Graph Convolutional Networks to integrate protein-protein interaction (PPI) networks with omics data.
• Low-dimensional embeddings for cell lines are constructed using integrated data.
• A deep neural network with Batch Normalization predicts synergy scores based on cell line and drug features.

Main Results:

• PRODeepSyn achieved a root mean square error of 15.08 and a Pearson correlation coefficient of 0.75, outperforming existing methods.
• In classification tasks, PRODeepSyn obtained an AUC of 0.90 and an AUPRC of 0.63.
• Ablation studies confirmed that integrating multi-omics data and PPI networks improves prediction performance.

Conclusions:

• PRODeepSyn effectively predicts synergistic anticancer drug combinations by integrating omics and PPI network data.
• The method demonstrates superior performance compared to existing deep learning and machine learning approaches.
• The findings highlight the value of incorporating biological network information into deep learning models for drug synergy prediction.