An MM-specific, transcriptome-based RAS classifier reveals genes driving the NRAS and KRAS phenotype. (A) Workflow for training and testing a gene-expression–based machine-learning algorithm to predict RAS genotype using an elastic net regression model. (B) Receiver operating characteristic curves for evaluating the performance of the predictive model on the training and testing sets. The area under the curve (AUC) is reported for each prediction class. (C) Confusion matrices showing the fraction of samples in each label-vs-predicted-class combination. (D) Multidimensional plot displaying weighted genes playing the most prominent role in predicting NRAS, KRAS, or WT RAS genotype.