Online Program

Return to main conference page
Friday, October 19
Fri, Oct 19, 2:30 PM - 3:45 PM
Caprice 3-4
Speed Session 4

Genomic Prediction of Tumor Location in Colon Cancer (304825)

*Emily Lora Herring, University of Notre Dame 
Steven Allen Buechler, University of Notre Dame  

Keywords: genomics, cancer, machine learning, classification

Background: Right-side colon cancer is shown to have worse prognosis than left-side cancers. Treatments have shown differential responses based on location as well. G2 is a two-gene classifier that predicts colon cancer tumor location. The predicted location is discordant with anatomical location for 15% of the samples. This project explores gene expression as it relates to tumor anatomical location. Methods: Machine learning methods were used on a gene expression dataset (n = 871) to determine if the G2 discordant samples were due to random error in the G2 classifier. A matched, tumor and normal tissue, dataset (n=54) was used to explore the nature of the discordance. Results: The discordant sample set from each of the classification methods overlapped significantly with that from the G2 classifier. In the matched dataset, discordance was higher in the tumor samples than in the normal samples, 18.5% and 7% respectively. Conclusions: Discordance is nonrandom and generally higher in tumor samples than in normal tissue samples of matched origin, suggesting it could be a trait acquired through carcinogenesis.