Predicting Drug Response Based on Multi-Omics Fusion and Graph Convolution.

Area of Science:

• Computational biology
• Genomics
• Pharmacogenomics

Background:

• Cancer patient responses to treatment vary due to tumor heterogeneity.
• Predicting drug response in cell lines is crucial for personalized cancer therapy.
• Existing computational methods require improvement for accuracy and scope.

Purpose of the Study:

• To develop an efficient end-to-end algorithm, MOFGCN, for predicting drug response in cancer cell lines.
• To leverage multi-omics data fusion and graph convolutional networks for enhanced prediction accuracy.
• To establish a foundation for personalized cancer treatment strategies.

Main Methods:

• Developed MOFGCN, an algorithm integrating Multi-Omics Fusion and Graph Convolution Network.
• Fused multiple omics data to compute cell line similarity.
• Constructed a heterogeneous network incorporating cell line similarity, drug similarity, and known cell line-drug associations.
• Employed graph convolution operations to learn latent features for cell lines and drugs.
• Utilized the linear correlation coefficient to reconstruct the cell line-drug correlation matrix for sensitivity prediction.

Main Results:

• MOFGCN demonstrated superior performance compared to state-of-the-art algorithms in predicting missing drug responses.
• Achieved higher accuracy in predicting drug responses for novel cell lines and drugs.
• Showcased effectiveness in predicting responses to targeted therapies.
• Validated through extensive experiments on the GDSC and CCLE databases.

Conclusions:

• MOFGCN represents a significant advancement in computational drug response prediction.
• The integration of multi-omics data and graph convolutional networks offers a powerful approach for personalized medicine.
• This method holds promise for guiding clinical decisions and improving cancer treatment outcomes.