KFSCIS Professors Ananda Mondal and Dongshen Luo recently published, “MOGAT: A Multi-Omics Integration Framework Using Graphic Tension Networks for Cancer Subtype Prediction in the prestigious International Journal of Molecular Sciences. Their work represents the first group to explore graph attention networks (GAT) in multi-omics integration for cancer subtype prediction.
To gain a comprehensive understanding of complex diseases like Alzheimer’s, Parkinson’s, and cancer, researchers have begun integrating multi-omics data into their work. Multi-omics allows researchers to incorporate data sets from multiple “omes” such as an organism’s genetic instructions or genome, a cell’s proteins or proteome, the set of RNA information or transcriptome, chemical composition or metabolome and microorganisms living on or within human tissue and bio fluids referred to as the microbiome. These studies provide the researchers a view into a disease or condition from a concerted approach to provide a comprehensive understanding of the disease. To enhance discovery of connections within these areas, numerous models have been suggested using graph-based learning to uncover veiled representations and network formations unique to a particular biological condition. Unfortunately, the existing approaches to identify cancer subtypes using graph convolutional networks (GCN) fail to “consider the level of importance of neighboring nodes on a particular node.”
Dr. Mondal’s research group explored the use of the graph attention network (GAT) to develop their multi-omics integration framework, MOGAT. In their research, they explored two sets of breast-cancer data using their new approach. Their findings demonstrated “that GAT embeddings provide a better prognosis in differentiating the high-risk group from the low risk group” in providing cancer subtype predictions in different scenarios.
To understand more about this seminal work, go to: https://www.mdpi.com/1422-0067/25/5/2788.