Figure 3.

Relationship between connectivity score and estimated tumour purity for 3 drugs previously developed and/or identified as HIF-inhibitors. (A) Average connectivity score (LOESS smoother) versus estimated tumour purity. At <80% estimated tumour purity the connectivity scores for CAY-10585 and PX-12 remain consistently negative, whereas YC-1 remains neutral across the range. (B) The same data for CAY-10585 and PX-12 but plotted separately for each drug across each estimated tumour purity tertile. The increase in negative connectivity from <60% tumour purity to 60–72% tumour purity is positive but statistically non-significant for CAY-10585 (P = 0.2) and statistically significant for PX-12 (P < 0.001). However, the decrease in negative connectivity from 60–72% tumour purity to >72% tumour purity is highly statistically significant for both (P < 0.001). Associations were tested with Wilcoxon rank-sum test on the connectivity scores. (C) Connectivity scores of PX-12 plotted against the connectivity scores of CAY-10585. The connectivity scores of CAY-10585 and PX-12 correlate well (Spearman Rho = 0.56, P < 10⁻¹⁶). (D) Distribution of tumour purity for 2 groups with different probabilities of HIF-activation. Samples were put into the “More likely” group if it produced negative connectivity scores with both CAY-10585 and PX-12 (i.e. in the bottom left of (C), and into the “Less likely” group if otherwise. Samples in the “Less likely” HIF-activation group have a higher estimated tumour purity on average (P = 0.006, Wilcoxon rank-sum test). From this figure it is easy to see that at higher tumour purity (i.e. >80%) samples are relatively are increasingly less likely to show HIF-activation.