Figure 8

Implementation of the training phase using the GA. Twenty splits of the training data set are considered where each split is a collection of 2000 randomly selected predictions that include 75% of the hits from the generated data for each protein complex. Training is repeated 10 times on each split and the best set of weights from each split are recorded. The set of weights that gives the best average BEDROC on the entire training data set is applied to the test set predictions.