Breast Cancer Classification via Graph Convolutional Networks with Attention Mechanism Utilizing Multi-Omics Data and Feature Selection Methodology

Abstract

This study emphasizes the integration of clinical data, Copy Number Alteration (CNA), and gene expression data to present an impactful methodology for the classification of PAM-50 breast cancer subtypes. Since breast cancer is a diverse disease, identifying its subtypes with precision is essential to developing therapies tailored to individual treat­ment plans. Given the variety of molecular traits that contribute to the complexity of breast cancer, this work is relevant because it tackles the problem of using multi-omics data to improve subtype classification. We commit to the inclusion of informative fea­tures by using Boruta for feature selection on single-omics data. Graph Convolutional Networks (GCN) help us to capture complex relationships and dependencies within the multi-omics dataset by integrating these various data modalities. This work is important not just because of its methodology but also because it advances precision medicine and cancer research in general. By increasing the precision of PAM-50 sub­type classification, the suggested method may help physicians make better-informed choices about treatment plans. The integration of multi-omics data for a thorough understanding of breast cancer might have advanced with this work, which empha­sizes the significance of taking clinical, genomic, and expression data into account simultaneously when characterizing subtypes.