Fig. 3: Analysis of drug response prediction in prostate cancer (PCa) cell lines.

a Heatmap showing predicted LN IC50 (Z-score) for 155 drugs across five PCa baseline cell lines highlighting PI3K/mTOR signaling targeting drugs. The lower the LN IC50, the more sensitive a sample is predicted to a drug. Color bars indicate different cell lines. Euclidean distance was used for grouping cell lines. b Ridgeplot showing the overall distribution of predicted LN IC50 (Z-score) across five PCa cell lines, while pooling their predicted drug sensitivities across biological replicates. c Scatterplot depicting Pearson correlation (ρ) between observed and predicted LN IC50 for the LNCaP cell line. The line color indicated two biological replicates of the LNCaP cell line. P-value was calculated using a two-sided t-test. Shaded areas depict a 95% confidence interval. d Heatmap of predicted LN IC50 (Z-score) of LNCaP cells in the presence and absence of androgens (DHT) and AR antagonists (ENZ, BIC, and APA) to 155 drugs. Euclidean distance was used for grouping samples. e Boxplots depicting the distribution of GSVA scores of proliferation-related pathways in the presence and absence of DHT. Notably, n = 12 pathways and n = 48 samples originating from n = 8 treatment groups (DHT, BIC.DHT, ENZ.DHT, APA.DHT, VEH, BIC.VEH, ENZ.VEH and APA.VEH) have been considered for this analysis. P-values were obtained from the two-sided Wilcoxon rank-sum test. f Boxplot depicting predicted LN IC50 for DNA replication targeting drugs (n = 15) across all treatment conditions (n = 8). Cisplatin is denoted using darkred colored filled triangle and other drugs are represented using grey filled circles. P-values were obtained from the two-sided Wilcoxon rank-sum test. g Boxplot showing the distribution of predicted LN IC50 values by n = 5 pre-trained models (based on cross-validation and hyperparameter tuning) for two drugs, metformin and orlistat. P-value was obtained by using a two-sided t-test. As expected, the direction of the relative difference in drug sensitivity is captured correctly even at the log scale. The structure of these two drugs, along with observed IC50 values is also depicted in h. In all boxplots (e, f, g), the middle horizontal line represents the median value. Each box spans the lower quartile to the upper quartile. The whiskers indicate the minimum and maximum values within 1.5 times the IQR. Source data are provided in the Source Data file.