Statistical analyses

R [61] and Python [62] were used for all statistical analyses. Confidence intervals at 95% probability for the tree root in P. virginalis are taken as the 95% highest posterior density (HPD) interval. All statistical tests were unpaired and two-sided, with a level of significance set at 5%. Segmentation p-values correspond to the test of significant difference between segments (Davies’ test, R package segmented). To check against overfitting, k-fold cross-validation of the segmentation procedure was conducted, with k = 10 folds. The resulting difference between test and validation mean square errors (MSE) was then examined. A non-increasing difference, when including one additional phase, indicates the absence of overfitting. Correlation coefficients between SBS1 and SBS5 were determined using Pearson method, and summarized by their median and IQR over GBM samples. A comparison between groups was made using an unpaired Wilcoxon rank-sum test. Differential time to recurrence between subgroups in the manually sorted ωγN curves was assessed using a Kruskal-Wallis rank-sum test between curve types "Peak", "Increase" and "Paused Start". The "Convex" curve type was excluded, because there was only 1 instance of this kind of curve, in the subset of samples with infomation on the time to recurrence. The association of the γ_R/γ_P ratio with the time to recurrence was assessed with a linear regression, using a simple or double log₁₀ scale on the γ_R/γ_P ratio, with Bonferroni adjustment.