Yes, Toolbox performs “leave-one-out” and Q2* is the result of this calculations. The procedure calculating Q2 is the following – each chemical from the training set is excluded from the set, the model is derived without taking into account the excluded chemical and then the excluded chemical is predicted with the new model. The Q2 value is the coefficient of determination in the plot “observed vs predicted” for all chemicals from the training set where the predictions are made by a model not including the predicted chemical in its training set.
Usually when count of analogues is bigger the procedure is slow and this is why it is not executed automatically, but only on demand.
Other methods of cross-validation are not implemented.
*Q2 is a measure obtained via “leave-one-out” method for the model predictivity over external chemicals (not included in the training set).