Extended Data Fig. 5: Drug-target connectivity identified by COSTAR.
a, Visualization of the local optimum in the cross-validated predictive power of COSTAR LASSO regression when fitting a binomial model to predict drug activity by PCY (hit vs neg) based on a drug’s connectivity pattern (COSTAR constellation, shown in Fig. 4b). X-axis denotes the Lambda regularization parameter (n = 60 unique values) and the y-axis denotes the cross-validated error of the model (deviance) across independent bootstrapped runs (n = 20 runs). Red dots (average) and light grey error bars (standard deviation) are indicated. Vertical dashed lines and colored circles indicate either the Lambda value with the minimal mean squared error (green, MSE) or the more conservative Lambda value with minimal MSE plus one standard deviation (blue, MSE + 1STD). b, ePTGs (x-axis) ranked by their integrated contribution ‘C’ to predict a hit (+1) or non-hit (-1) (y-axis) in the COSTAR model, separated for PCY-hit NADs (left) and ONCD (right). c, Drug-target connectivity of PCY-hit drugs that were part of the COSTAR training data (columns; n = 30 drugs) to primary and secondary drug targets (rows). COSTAR subscore (heatmap color scale) is the LASSO model coefficient multiplied by the integrated connectivity of drug-to-target mapping. Target genes with absolute COSTAR LASSO coefficients >0.1 are displayed. Target level (primary or secondary target) is annotated per gene on the left.
