Figure 4. p-RSK^Th573 is the most discriminative signal.

(A) Schematic showing the cross validation (CV) procedure used to train and test our linear regression models. For each signal, up to eight CV models were trained and tested. Pearson’s correlation coefficient (R_all), Kendalls’ correlation coefficient (τ_all), and root mean squared error (RMSE_all) were used to evaluate the models. (B) Scatter plots showing the measured versus predicted sensitivity levels of NSCLC cells (dots) for two of the CV models based on the p-RSK^Th573 (black = training, red = test cell lines). In these two CV models shown, the most sensitive (H460) or resistant (EKVX) cells were not used for model training, respectively. (C) Mean performance values of the CV models based on each or combinations of the seven selected signals. For each condition, the number of CV models that could be fitted is shown in the parentheses (*P < 0.05, **P < 0.01, ***P < 0.001, one-sided t-test.) (D) The average weights (β_i) of p-RSK^Th573 events across the eight CV models. Only weights with |β_i| > 0.03 are shown. F₁ and F₂ = two p-RSK^Th573 events with the highest |β_i| values. (E) The average values of F₁ (blue) and F₂ (green) in all the eight cell lines (error bars = SEM). (F) Mean performance values of the CV models based on p-RSK^Th573 events, genome-wide mRNA expression levels, DNA copy number variations, or combinations of these features of the cell lines (*P < 0.05, **P < 0.01, ***P < 0.001, one-sided t-test with respect to the CV model based on p-RSK^Th573 events).