Fig. 6

Comparison of the ratio accumulation curves for similarity of any molecular pairs for the same labels (positive–positive or negative–negative) in blue and the different labels (positive–negative or negative–positive) in red. The values of the kernel matrix after the optimization and the Tanimoto coefficient are used for the new descriptor and ECFP, respectively. a The training set of the MUTAG set used in Fig. 5. b A training set of the PTC-FM set