Figure 1

Splitting of training sets for GP algorithm. The figure shows three families from an example superfamily that constitute the positive training set for GP. The positive test set and the negative training and test set are not shown in this figure. In addition to train motifs to cover the sequences of all three families, we also train GP on all the possible subsets of this superfamily that exclude one family of the positive training set. This is indicated by the four sets in the right part of the figure. Ten motifs are made for each subset.