Introduction of an information-theoretic method to predict recovery rates of active compounds for Bayesian in silico screening: theory and screening trials.

We present the first method to predict compound recovery rates from descriptor statistics. A log-odds function is designed that models probability distributions of descriptor values of active and inactive molecules in chemical space and used to determine the likelihood of database compounds to exhibit a specific activity. The divergence of probability models for active and inactive compounds is applied to evaluate the ability of the log-odds likelihood function to recover active compounds from a background database. The divergence measure, which is closely related to the Kullback-Leibler distance, is strongly correlated with recovery rates of Bayesian virtual screening calculations. It has thus been possible to predict compound recovery rates for different activity classes. Prior to practical virtual screening trials, one can also estimate how likely it would be to recover active compounds from a given screening database.