Fig. 4
Overall approach for investigation of the effect of subsampling on NEAT predictive performance. A total of 106 combinations of n = {1,2,3} samples per tumour were analysed across the 22 patients in the validation cohort where multiregion sampling encompassed identified morphological intratumoural heterogeneity (top left). A median of four samples was taken per tumour. The distributions of logHR and log-rank p values across the 106 samples taken for each value of n (bottom right) are given at readable size in Fig. 5