Binning of Cold-Responsive and Nonresponsive Genes.

A binning method was used to reduce the bias of baseline gene expression and to balance the number of genes in the cold-responsive and nonresponsive datasets for supervised machine-learning classification. The joint set of all cold-responsive and nonresponsive genes was sorted and segmented into 12 bins (dodeciles) based on average expression value. Within each dodecile, all genes of the less abundant class (either cold-responsive or nonresponsive) were included as potential data points for training and testing, while the more abundant class was randomly subsampled to provide equal numbers of cold-responsive and nonresponsive genes within that particular dodecile.