Discussion of used machine learning algorithms

Naïve Bayes algorithm is based on the Bayes theorem and the assumption that particular attributes are conditionally independent. Obviously, this assumption in usually not true (therefore a classifier is called “naïve”). The algorithm calculates the probability of belonging to particular class independently; as the final answer, the class with the highest probability is selected. Therefore, the ROC curves for Naïve Bayes could provide slightly different information than it is in case of the other classifiers. Labeling the compounds as active or inactive is not based on the absolute value of probability of belonging to particular class but simply the class with the highest probability is chosen. That is why for example in one case, probability of being active equals to 0.6 is sufficient for a given compound to label it as active (if the probability of being inactive is lower), and there could also be cases that having the probability of 0.9, the compound is inactive (if the probability of being active was lower, e.g. 0.89).