Oncology researchers frequently seek to associate mega-dimensional omic data with clinical characteristics and outcomes such as response to chemotherapy, time to relapse, and time to death. Here, bootstrap evaluation of association matrices (BEAM) is introduced as a robust method that can be applied to a broad class of these types of research studies.
For each gene, BEAM first computes a matrix of estimates of the association of each clinical outcome with each molecular variable using the observed data. Next, BEAM repeats these calculations for each of many bootstrap data sets obtained by resampling subjects. These calculations yield a cloud of association estimates in multivariate space. Finally, to quantify statistical significance, BEAM uses a recursive peeling algorithm to compute a layer value (l-value) that characterizes the position of the null vector relative to the observed association estimate vector and the cloud of bootstrap association vector estimates.
Simulation studies and example analyses of pediatric leukemia data sets indicate that BEAM can very effectively identify clinically important genes.