 This paper examines the effect of institutional bias on deep learning models trained on the TCGA dataset. The authors found that certain institutions had a greater influence on the model's performance than others, suggesting that the models may not be generalizable across different institutions. Additionally, they discovered that the models were able to pick up on institution-specific patterns which could potentially interfere with other applications of deep learning in digital pathology, such as image search. This article was authored by Tohei Dekarganian, Azama Sillian Bidgeli, Actin Ryosaitian, and others.