Proteins play a fundamental functional role in the molecular processes that underlie various diseases, including cancer. As a result, there is great interest both in studying proteins directly and in identifying proteomic biomarkers of cancer that can potentially be used for early detection, new drug target identification and precision medicine strategies.

But given the interpatient heterogeneity that is a hallmark of cancer, many potentially useful proteomic biomarkers may have aberrant expression in only a small subset of cancer patients—and these differences are not apparent when using standard methods. This team developed new methods to perform quantile regression for functional responses, which allowed them to more thoroughly assess how predictors affect the distribution of outcome measures. These fully Bayesian methods to perform quantile regression on functional responses enabled them to identify for which functional locations the distribution of measurements varies according to a particular covariate.

Their method outperforms existing standard results. In a pancreatic cancer proteomics biomarker study, the method was able to identify proteomic biomarkers that are only different in a subset of cancer patients—differences from controls that standard mean regression approaches missed.

Read the study in The Annals of Applied Statistics.