Validation of Biomarkers Predictive of Tumor Location in Coloadenocarcinoma, an Analysis of the TCGA COAD Dataset

Tumor localization correlates with prognosis in coloadenocarcinoma, with aboral tumors having a better overall survival. This can be attributed to their better response to biologicals such as the anti-EGFR (epidermal growth factor receptor) cetuximab. It was hypothesized that –regardless of the causal relationships– this “sidedness” of coloadenocarcinomas could be reconstructed on a genomic and transcriptomic level.

A two machine learning models, logistic regression and random forest, were implemented in python3 using pandas and sklearn, among other libraries.

There are two versions of the analysis, the original jupyter notebook, and a reworked script. The script uses the harmonized gdc api, can be configured using config flags, and handles caching more elegantly, and so should be preferred. However, paper.pdf references the jupyter notebook.