Root-mean-squared error (RMSE) comparison among different models on the connectivity map gene expression data.

<p>“Lasso” represents model (2), “Linear low-rank” represents model (3), and “Nonlinear low-rank” represents model (6) with ANOVA kernel. Both training and testing errors are measured in terms of RMSE averaged over all target genes. Shown here are mean ± standard deviation values of RMSEs in ten different runs.