[Synergistic drug combination prediction in multi-input neural network]

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug combinations enhance efficacy and reduce toxicity, but screening is costly.
• Virtual screening using computational models can reduce experimental costs.
• Existing deep learning models like DeepSynergy have limitations in feature integration.

Purpose of the Study:

• To develop an end-to-end deep learning model for predicting drug synergy.
• To integrate multiple data types for improved prediction accuracy.
• To overcome the limitations of existing two-stage, single-feature models.

Main Methods:

• Proposed MulinputSynergy, an end-to-end deep learning model.
• Integrated gene expression, mutation, copy number, and drug chemistry features.
• Utilized convolutional neural networks for gene feature dimension reduction.

Main Results:

• MulinputSynergy demonstrated superior performance compared to DeepSynergy.
• Mean squared error decreased from 197 to 176.
• Mean absolute error decreased from 9.48 to 8.77, and decision coefficient increased from 0.53 to 0.58.

Conclusions:

• MulinputSynergy effectively learns relationships between drugs and cell lines from diverse features.
• The model enables rapid and accurate identification of effective drug combinations.
• This approach holds promise for accelerating cancer drug discovery.