Graph Transformer for Drug Response Prediction.

Area of Science:

• Computational Biology
• Pharmacology
• Machine Learning

Background:

• Graph representations enhance drug feature learning over string or numeric data.
• Integrating multi-omics data improves drug response prediction accuracy.
• Existing models struggle with efficient graph-based drug feature extraction and multi-omics data redundancy.

Purpose of the Study:

• To propose GraTransDRP, a novel deep learning model for enhanced drug representation and reduced information redundancy in drug response prediction.
• To improve the efficiency of extracting drug features from graph representations.
• To effectively incorporate multi-omics data while minimizing redundancy.

Main Methods:

• Utilized Graph Transformer for efficient drug feature extraction.
• Employed Convolutional Neural Networks (CNNs) for mutation, methylation, and transcriptomics data.
• Applied Kernel Principal Component Analysis (KernelPCA) to reduce transcriptomics feature dimensionality before CNN processing.
• Integrated drug and processed omics features using a fully connected network for response prediction.

Main Results:

• The proposed GraTransDRP model demonstrated superior performance compared to state-of-the-art methods like GraphDRP and GraOmicDRP.
• Efficient extraction of drug features from graph representations was achieved.
• Effective integration and dimensionality reduction of multi-omics data improved prediction accuracy.

Conclusions:

• GraTransDRP offers a more effective approach to drug response prediction by optimizing drug representation and multi-omics data integration.
• The model addresses limitations of previous methods in handling graph-based drug features and redundant omics information.
• This work advances the application of deep learning in precision medicine for predicting drug efficacy.