Biomarker selection for omics data

The R package BioMark provides several tools to define which variables are associated with class differences in data from fields like metabolomics and proteomics. The first group of tools uses Higher Criticism to define an optimal threshold between interesting and non-interesting variables. This can be applied to any statistic, be it a t value, a regression coefficient or something else, and is related to the expected distribution of p values under the null distribution. The second group of tools is based on stability selection, i.e. an assessment of how often specific variables are highlighted as interesting under perturbation of the data. This approach is especially attractive when the number of samples is larger than, say, ten per group. Also in this case, the strategy can be applied to any type of statistic. Using real and simulated data, the application and usefulness of these techniques will be shown