Applying Machine Learning in Cancer Prognosis Using Expression Profiles of Candidate Genes

Cancer progression is a dynamic process that involves a wide spectrum of changes in expression levels for multiple genes. Increasing amount of data has been collected for patients, such as genome, transcriptome, prognosis, and histology images of the tumor. While previous studies have used mostly mutation profiles miRNA and ctRNA for cancer prognosis, we are interested in evaluating the use of expression levels of candidate genes in cancer prognosis. We used gene expression of 45 candidate genes in breast cancer as an example and showed the effectiveness of using such multivariate expression data in cancer prognosis to predict 5-year survivorship. Applying machine learning techniques, we were able to predict survivorship from expression data alone without incorporating any other information. Our example study urges such expression data to be collected for patients by hospitals, research institutes in the future.