Hierarchical classification treeAbstract
Metastasis is major contributor towards cancer-related mortality and can be difficult to detect during early stages. The ability to identify cancers that may have already metastasized can help increase patient survival. In this study, we utilize publicly available expression profile datasets of cancers from primary sites with or without distal metastasis. We train an elastic net models to predict the origin of primary cancer tissue and whether the primary cancer has metastasized or not. Using the elastic-net for hierarchical classification, we were able to predict the origin tissue at an accuracy of 97% and whether the cancer has already metastasized at an accuracy of 90%. When examining the top influential genes in the model we find that many mitochondrial genes were negatively correlated with metastasis.
