Fig. 2
From: DualGCN: a dual graph convolutional network model to predict cancer drug response
Performance of DualGCN across cancers and drugs. a Pearson’s correlation on each type of cancer. We calculated the average Pearson’s correlation coefficients of samples belonging to each type of cancer and sorted the coefficients from large to small (from left to right in the figure). Blue dots indicate the mean of Pearson’s correlation across CVs and are denoted by \(\overline{r}\). Vertical blue bars represent variances of Pearson’s correlation across CVs. \(\overline{n}\) denotes average sample size across CVs. The largest and smallest Pearson’s correlation coefficients were obtained on lung squamous cell carcinoma (LUSC) and neuroblastoma (NB), respectively. b Scatterplot of correlations between true and predicted IC50 on LUSC. c Scatterplot of correlations between true and predicted IC50 on NB. d Pearson’s correlation on each drug. We calculated the average Pearson’s correlation coefficients of samples belonging to each drug and sorted the coefficients from large to small. The left ten in the figure are drugs with the best predictive performance, and the right ten are drugs with the worst predictive performance. Blue dots indicate the mean of Pearson’s correlation across CVs and are denoted by \(\overline{r}\). Vertical blue bars represent variances of Pearson’s correlation across CVs. \(\overline{n}\) denotes average sample size across CVs. The largest and smallest Pearson’s correlation coefficients were obtained on CAY10603 and cetuximab, respectively. e Scatterplot of correlations between true and predicted IC50 on CAY10603. f Scatterplot of correlations between true and predicted IC50 on cetuximab