When predicting protein structures, normally many different possible structures are made. These candidate conformations are known as decoys. By making many decoys (generating them computationally), the probability that one of them closely resembles the true structure of the target protein is increased - however, since the true structure is unknown, a method must be developed that can predict which of the decoys is the most accurate. One way of doing this is via the use of a statistical (or knowledge-based) energy function, known as Statistical potential