Fig. 2.

Cross-validation performance of gene-expression biomarker with different quantities of genes. For the gene-expression biomarker, we used the SVM-RFE method to identify genes whose expression differed most consistently between individuals who developed familial breast cancer and individuals who did not. The sizes of these gene subsets ranged in size between 25 and 300 genes. In repeated cross-validation (1,000 iterations), predictive accuracy peaked at 250 genes and was consistent when the number of genes was 150 or higher