DUPROSY: Dual probabilistic system for biochemical activity prediction

Various methodologies have been proposed for in-silico prediction of biochemical properties: all reporting similar recognition rates but with individual strengths and optimality conditions. In this submission a dynamic strategy is proposed: depending on the input compound, it is decided which model should be used for prediction. In the dual probabilistic system DUPROSY for prediction of biochemical activity, complementary methods are given a probabilistic interpretation and joined via a probabilistic graph. In questionable cases, where there is no unanimous answer by the constituent predictive models, a locally suitable predictive model overtakes the decision process. Testing results prove the DUPROSY to yield more reliable predictions than those by the state of the art individual constituent methods.