On statistical analysis and machine learning in prostate cancer research

The machine-learning age opens new opportunities for data analysis in multiple contexts spanning from business analytics and traffic control to weather forecasting and natural language processing. However, there might be untapped potential in biological and biomedical basic, clinical, andobservational research to apply these relatively new methodologies. In biomedicine, the research objectives are typically explored by classic frequentist statistics or descriptively explored whereas machine learning methodologies might still get overlooked. That is, the state-of-the-art in the analysis of biomedical data has not shifted towards machine learning. We explored the possibilities of machine learning in prostate cancer research and compared the results with classic frequentist method. In our analysis, machine learning was represented by random forests and classic frequentist method by Wilcoxon rank sum test, which were applied into same analysis problem and high-dimensional datasets. The inferences from these methods were concordant, that is, a researcher could arrive at same conclusion by using either method. While machine learning is still nascent technology, this suggests that it is ready to use in biomedical cancer research; however, whether biomedical cancer research is ready for machine learning remains to be seen.