Fig. 8: Relationship between genomic features and patient outcome.

a,b, Kaplan–Meier curve on PFS (a) and OS (b) of TP53 altered/WT in combination with GC7/8. The P value was derived from a log-rank test comparing the most two extreme curves (additional data in Extended Data Fig. 10). The dotted lines indicate the median survival for each subgroup. c,d, Genomic factors comprising the GS (cut-off 0.5) derived using non-negative matrix factorization, hypermutated subset (u-GS) (a), unmutated subset (m-GS) (b). The plot only shows features that split the data. e,f, Kaplan–Meier curves of PFS of samples divided by GS. Only samples with PFS data were included (n = 243). In e, the unmutated subset, del17p/TP53 mutated samples are plotted separately (black curve), all u-GS1 cluster 1 samples fell into this grouping; In f, the hypermutated samples, del17p/TP53 mutated samples are plotted separately (black curve). The P value was derived from a log-rank test comparing the most two extreme curves. The dotted lines indicate the median survival for each subgroup g, Confusion matrix showing agreement between true and predicted subgroup assignment. The true subgroup assignment was determined by applying the previously described NMF approach (Methods) to the whole set of genomic data. The predicted subgroup assignment was determined by first using 80% of the genomic data for subgroup assignment (training phase) followed by predicting the subgroup assignment in the remaining 20% of the data (testing). In all cases, sex and age were included to inform the model (Methods).