Fig. 3

Performance scores under varying number of informative genes. Performance scores and 99%-confidence bands for a “expected fraction of false findings in signature" FDR, b “expected error on single future predictions” MSPE, c “chance of successful validation" SV, and d “average calibration slope” CS calculated from 10³ simulated prognostic modeling iterations. The parameter values are given in Table 1 and the number of informative genes is varied. Note that the overall signal (\(\sum _{g} \tilde {\beta }_{g} \equiv \beta \cdot n_{inf}\)) is kept constant by adapting \(\tilde {\beta }\) to the number of informative features n_inf