Fig. 2

The landscape of LncRNA-drug predictive pairs in cancer cell lines. a Effect of cell lineage on drug response prediction for each agent. The linage effect is evaluated using one-way ANOVA, and negative log10-transformed p-values are indicated on y-axis. The agents are organized based on targeting pathways (x-axis). b Volcano plot of pan-cancer models (left) and cancer-specific models (right) performance in drug response prediction in the bootstrapping process. Pearson correlation coefficients (x-axis) and negative log10-transformed p-values (y-axis) indicate the model performance. c lncRNA-drug predictive pairs landscape across 265 agents and 505 cancer cell lines. The predictive score for each lncRNA-drug interaction and the negative log-transformed p-value for Pearson’s correlation between the lncRNA expression and IC50 were shown in the y-axis and x-axis of the volcano plot. d The distribution of Spearman’s correlation coefficients between lncRNA expression and ln-transformed IC50s. The density plot of the coefficients is shown for (i) strong predictive pairs with PS >= 0.8; (ii) moderate predictive pairs with 0.25 <= PS < 0.8; (iii) weak predictive pairs with 0 <= PS < 0.25; (iv) non-predictive pairs and (v) the combination of all lncRNA-drug pairs. e Scatter plot of the predictive score between lncRNA-drug predictive pairs identified in IC50 models and those identified in AUC models. f Agents targeting genome integrity clustered by shared predictive lncRNA signatures. One-sided Fisher exact test p-values were indicated by different colors in the heatmap