Fig. 8 | Nature Communications

Fig. 8

From: Community assessment to advance computational prediction of cancer drug combinations in a pharmacogenomic screen

Fig. 8

Translatability of AstraZeneca-DREAM models to the independent screens. Performance of SC1A models for predicting synergy scores in the O'Neil et al.4 dataset by the best performing teams are plotted along with distributions of predictions from the random model and replicate experiments. Performance of predictions are shown for a all experiments in the O’Neil et al.4 dataset, and three subsets of the dataset; b experiments that tested same cell lines as AZ-DREAM, c tested similar drugs as in AZ-DREAM (one drug in the combination with the same target), and d tested similar combinations as in AZ-DREAM (same targets for both drugs in the combination). e Prediction performance on the ALMANAC13 dataset is shown for a random, NAD baseline (using only cell line and drug identities as features) and full NAD model for no targeted agents (None, i.e., two chemotherapeutic drugs), one targeted agent plus chemo (One drug), and combinations of two targeted agents (Both drugs)(mean +/− 95% CI). ANOVA P-values: *P = 1.322e-31 for different performance between baseline and full model; #: P = 5.584e-5 for association between model performance and number of targeted drugs in the combinations

Back to article page