Figure 3

Pre-processing influences biomarker validation. (a) Results for all Cox proportional hazard ratio modeling analysis for the 24 different pre-processing schemes in the Director's Challenge dataset [21] are summarized in Forest plots. (b) Classifications in the 24 different schedules are visualized in a heatmap. To confirm that biomarker performance and individual patient classification are highly dependent on dataset pre-processing, all pre-processing schedules were tested in a second dataset [29] for the three-gene biomarker. (c, d) Both biomarker performance (c) and individual patient classifications (d) were again influenced by differences in pre-processing. For the Forest plots; boxes and lines are the hazard ratios and 95% confidence intervals, respectively. For the heatmaps; white indicates a patient predicted to have good prognosis and black indicates a patient predicted to have poor prognosis. Colored sidebar displays the different pre-processing schemes as explained in the legends.