Multi-omics data integration by generative adversarial network.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Accurate disease phenotype prediction is crucial for precision medicine, especially in heterogeneous diseases like cancer.
• Multi-omics data offers comprehensive genotype-phenotype links but presents integration challenges due to interactive biological layers.
• Discovering coherent biological signatures and predicting phenotypic outcomes from multi-omics data remains a significant challenge.

Purpose of the Study:

• To introduce omicsGAN, a generative adversarial network model designed for integrating two omics data types and their interaction network.
• To capture and fuse information from interaction networks and omics datasets to generate synthetic data with enhanced predictive capabilities.
• To improve the prediction of cancer outcomes and patient survival by leveraging integrated multi-omics information.

Main Methods:

• Development of omicsGAN, a generative adversarial network model.
• Integration of two omics data types (e.g., mRNA and microRNA expression data) with their corresponding interaction network (e.g., microRNA-mRNA interaction network).
• Generation of synthetic omics data using the omicsGAN model.

Main Results:

• omicsGAN effectively integrates multi-omics data and interaction networks, demonstrated on The Cancer Genome Atlas (TCGA) breast, lung, and ovarian cancer datasets.
• Synthetic omics data generated by omicsGAN showed superior performance in cancer outcome classification and patient survival prediction compared to original datasets.
• The integrity of the interaction network is critical; random networks yielded synthetic data with weaker predictive signals, highlighting the importance of biological network information.

Conclusions:

• omicsGAN provides a robust framework for integrating multi-omics data and biological networks to improve disease phenotype prediction.
• The model's ability to generate high-predictive synthetic data offers a promising approach for advancing precision medicine in oncology.
• The study underscores the vital role of biological interaction networks in enhancing the predictive power of integrated multi-omics analyses.