In recent years, a great deal of biomedical research has been focused on identifying biomarkers that can be used in settings of clinical research and practice to evaluate exposure, effect, or susceptibility of a patient to external stimuli. One such area of research has sought to discover and classify biomarkers that can be used to guide treatment selection in patients, especially those with different types of cancer. Despite various attempts, there remains a lack of consensus about how to best objectively select candidate genes that could inform medical decisions to maximize the treatment outcome for patients. The purpose of this study was to investigate two common statistical methods used for prediction, and to compare them under simulated circumstances to evaluate their internal and external validity in a real-world clinical setting. Overall, the lasso regression model displayed a greater robustness to various levels of simulated variation compared to the neural network.