Fig. 12

a Cavities A, B and C have the same binding preferences with subtle steric differences. b To produce training data, all symmetric CSG differences are computed between the training set cavities. c Individual fragments (semicircles) are separated from each CSG difference, and their volumes are computed. A log-normal probability density function is fitted to the volume data. d To estimate the probability that a given cavity D has similar or different binding preferences, a solid representation of the cavity is first generated (yellow) and symmetric CSG differences are computed between D and a randomly selected training set protein C. e The largest fragment found in the CSG differences \(C-D\) and \(D-C\), is shown in blue. f The log normal distribution enables the p-value to be estimated resulting in high probabilities for small fragments and low probabilities for large fragments