Progressively Helical Multi-Omics Data Fusion GCN and Its Application in Lung Adenocarcinoma

Abstract

Compared to single-omics data, utilizing multi-omics data helps to gain a more comprehensive understanding of the occurrence and development of cancer, which emphasizes the necessity of developing efficient multi-omics data fusion approaches. In this study, a novel framework based on graph convolution neural networks with a progressively helical multi-omics data fusion strategy, named phMFGCN, is proposed to effectively integrate multiple omics data. To demonstrate the effectiveness of our framework in addressing the challenges of multi omics data fusion, phMFGCN and other widely-used machine learning methods conducted comparative experiments on predicting gene-gene interactions in lung adenocarcinoma. The results illustrated that phMFGCN outperforms other models with an accuracy of 97.94%. Additionally, 506 new gene-gene interactions predicted by this framework have been validated in databases such as BioGrid. Finally, it was used to perform gene function prediction, and the results were inconsistent with other existing research, for examples: Sam68, DHX9, and HNRNPK were involved in regulating multiple lung adenocarcinoma related pathways simultaneously. All these results demonstrate the universality of phMFGCN for different clinical tasks and it can provide reference target genes or gene-gene interactions for cancer mechanism research and treatment research in clinical practice.

Keywords

• Multi-omics data fusion
• graph convolution neural network
• lung adenocarcinoma
• gene-gene interaction