Cancer classification has progressively shifted from histopathological assessment to molecular profiling, particularly RNA-seq gene expression and, more recently, multi- omics integration approaches. Most methods rely on static gene features or predefined interaction networks and rarely incorporate gene regulatory structure. In addition, cross- cancer disease association remain largely disconnected from individualized gene regulatory modeling. This thesis addresses this gap by integrating patient specific gene regulatory networks -inferred via LIONESS- with graph neural network models, exploring an under- investigated space at the intersection of multi-omics integration, network biology, and cross-disease cancer modeling. The origin datasets are retrieved from the Genomic Data Commons (GDC) portal to construct a large, curated RNA-seq cohort comprising leukemia, breast cancer and normal controls.