Hypothesis: Through prediction of chromosomal instability genes different gene expression prediction models trained on WSIs under the same benchmark metrics leads to comparable results and these estimates are significantly associated with patient survival outcomes and improve prognostic stratification beyond standard histological grading alone. Background: Histological images are a valuable tool for pathologists to diagnose cancer as a qualitative prognostic marker. Yet the differences between professionals can lead to non-homogenous tumor grading. RNA data can be used then to better grade and diagnose the disease. Paired with a biomarker such as chromosomal instability, bulk RNA is a powerful tool not only for diagnosis, but also for treatment options. But it is not as readily available as a histological image is. This is where deep learning models predicting bulk RNA from histological whole slide images come in. Models such as SEQUOIA, HE2RNA, RNAPath and many others hold promise as a valuable prognosis tool in cancer. Nevertheless, due to a lack of benchmarking background within these models, comparisons between each other can be lacking. In this project, we aim to benchmark such models in TCGA breast cancer data specifically using the chromosomal instability signature genes and using this biomarker as a predictor for survival outcome in breast cancer patients. Methods: SEQUOIA and HE2RNA are open-source Deep learning models readily available in GitHub and trained in TCGA histological images for breast cancer prediction. These will be implemented and ran with TCGA BRCA data to compare gene prediction using the 70 genes related to chromosomal instability signature (CIN70). Through patch-level prediction, we will obtain spatial heterogeneity of the CIN70 signature on the histological images. On the other hand, RNAPath is a visual transformer trained to predict spatial transcriptomics. We will compare it with SEQUOIA and HE2RNA.