Synergistic Drug Combination Prediction by Integrating Multiomics Data in Deep Learning Models

Area of Science:

• Computational biology
• Genomics
• Drug discovery

Background:

• Drug resistance remains a significant hurdle in effective cancer treatment.
• Identifying synergistic drug combinations offers a promising strategy to overcome resistance.
• The vast number of potential drug combinations makes experimental screening infeasible.

Purpose of the Study:

• To develop a novel deep learning model, AuDNNsynergy, for predicting the synergy of pairwise drug combinations.
• To integrate multiomics data (gene expression, copy number, genetic mutations) for enhanced prediction accuracy.
• To computationally prioritize effective drug combinations for cancer therapy.

Main Methods:

• Trained three autoencoders on The Cancer Genome Atlas (TCGA) multiomics data (gene expression, copy number, mutation).
• Encoded cancer cell line omics data using the trained autoencoders.
• Developed a deep neural network integrating encoded omics data and drug physicochemical features to predict synergy scores.

Main Results:

• The AuDNNsynergy model demonstrated superior performance in predicting drug combination synergy.
• Outperformed four state-of-the-art methods, including DeepSynergy, Gradient Boosting Machines, Random Forests, and Elastic Nets.
• Achieved higher accuracy, particularly in the rank correlation metric, for prioritizing effective drug combinations.

Conclusions:

• AuDNNsynergy provides an effective computational approach for predicting synergistic drug combinations.
• The integration of multiomics data significantly enhances the prediction of drug synergy.
• This model can accelerate the discovery of novel combination therapies to combat cancer drug resistance.