Prediction of anticancer drug sensitivity using an interpretable model guided by deep learning.

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Predicting drug sensitivity is vital for effective cancer therapy.
• Current methods lack interpretability and struggle with complex drug reaction mechanisms.
• There is a need for interpretable models using diverse cell line and drug data.

Purpose of the Study:

• To develop an interpretable deep learning model for predicting anticancer drug sensitivity.
• To integrate multi-omics data from cancer cell lines with drug chemical features.
• To understand drug response mechanisms and improve prediction stability.

Main Methods:

• Proposed DrugGene, an interpretable deep learning model.
• Integrated gene expression, mutation, copy number variation, and drug chemical structures.
• Employed a visual neural network (VNN) for cell line genotype analysis and an artificial neural network (ANN) for drug features.
• Combined VNN and ANN outputs for final drug response predictions.

Main Results:

• DrugGene demonstrated superior performance compared to existing prediction methods.
• The model accurately predicts drug sensitivity by learning reaction mechanisms.
• Achieved higher accuracy and provided interpretable prediction results.

Conclusions:

• The developed model utilizes biological pathways to construct interpretable neural networks.
• Genotypes are used to monitor subsystem states, enabling interpretation of predictions.
• The approach offers satisfactory prediction accuracy and aids in exploring new cancer treatment strategies.