MoSwformer: a transformer-based patient classification model using multiomics data

Cancer is a global health challenge, and the utilization of multi-omics data offers a comprehensive approach to understanding and predicting human disease. Despite the abundance of gene expression, DNA methylation, and protein-protein interaction data, the integration of these datasets for accurate cancer classification prediction remains challenging. This paper introduces MoSwformer, a novel multi-omics framework for multi-omics analysis that combines Transformer based on sliding window embedding (SET), multi-omics attention mechanisms and omics integrated representation learning (MAL). By leveraging SET for feature extraction and MAL for multi-omics data fusion, MoSwformer improves cancer subtype classification. The framework serves as a powerful tool for multi-omics data analysis, highlighting its potential in advancing cancer research and precision medicine. Additionally, the findings from the ablation study provide convincing evidence for the indispensable role of the three integral modules within the MoSwformer framework. MoSwformer is freely available at https://github.com/ShiLab-GitHub/MoSwformer.