DNN-PNN: A parallel deep neural network model to improve anticancer drug sensitivity

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Deep learning and large-scale genomics databases offer potential for predicting anticancer drug sensitivity.
• Accurate prediction can improve the identification of therapeutic biomarkers.

Purpose of the Study:

• To propose a novel parallel deep learning framework, DNN-PNN, for anticancer drug sensitivity prediction.
• To introduce an effective data representation strategy for high-dimensional discrete data.
• To optimize the framework for reduced time complexity.

Main Methods:

• Developed a parallel deep learning framework (DNN-PNN) integrating gene expression and chemical structure data.
• Introduced a new feature correlation strategy using product representation.
• Optimized the framework for computational efficiency.
• Conducted experiments on CCLE datasets comparing DNN-PNN with other models.

Main Results:

• DNN-PNN demonstrated high prediction accuracy for anticancer drug sensitivity.
• The framework showed significant advantages in model stability and convergence speed compared to baseline models.
• The product-based feature representation alleviated limitations of high-dimensional discrete data.

Conclusions:

• DNN-PNN offers a powerful and efficient approach for predicting anticancer drug sensitivity.
• The framework advances the application of deep learning in precision oncology.
• This method can enhance the discovery of effective therapeutic biomarkers.