Deep Neural Network-Assisted Multi-Dimensional Genomic Data Integration and Disease-Associated Gene Discovery

The advent of high-throughput sequencing technology has led to the generation of a proliferation of multidimensional genomic data, including, but not limited to, transcriptome, epigenome, and genetic variation. However, due to the heterogeneity and high dimensionality of the data, there is a paucity of research on the effective integration of these complex data and accurate identification of the genes associated with diseases. In this paper, we propose a novel framework based on deep neural networks (DNNs) for the integration of multidimensional genomic data, including gene expression, DNA methylation, and genetic variation, and for the discovery of disease-related genes. The proposed method is founded on existing multi-view learning and graph neural network models, integrating advanced technologies such as the attention mechanism and multi-task learning to enhance the data fusion process and prediction accuracy. The framework employs a multi-layered architecture that first learns representations from each data modality. Furthermore, a novel regularization strategy has been introduced with the aim of preventing overfitting and more accurately capturing the complex associations between genes and diseases. The experimental results showed that compared with existing state-of-the-art methods, the proposed model significantly improved the recognition ability of disease-related genes by effectively integrating complementary information from multiple genomic data sources.