Fig. 4

Multivariate predictive models for response to anti-CTLA-4 treatment. Multivariate analysis based on the FlowJo gating of CyTOF data was performed to generate a model that predicts clinical response to anti-CTLA-4 monotherapy. The heatmap is row-normalized and shows the four features used by the model with the highest area under the curve (AUC). Each value in the heatmap corresponds to a cell population as determined by manual gating, where red indicates a larger population and blue a smaller one. The signature coefficients (grey bars) on the left side indicate the elastic net weights, which corresponds to the importance in the prediction model. The clinical response of each patient is indicated in green (responder) and red (non-responder) as a horizontal bar above the heat map. The predictions of clinical response for each patient are indicated in grey (responders) and black (non-responders), where the height of the vertical bar represents the probability output of the predictor. The receiver operating characteristic (ROC) curve for the model is shown on the right