Fig. 1. Illustration of the procedure for inferring positive selection.

The method includes two parts. Part I (left) is the gapped k-mer support vector machine (gkm-SVM) model training. The gkm-SVM classifier was trained by using TFBSs as a positive training set and randomly sampled sequences from the genome as a negative training set. Then, SVM weights of all possible 10-mers, the contributions of prediction transcription factor binding affinity, were generated from the gkm-SVM. Part II (right) is the positive selection inference. The ancestor sequence was inferred from sequence alignment with a sister species (species B) and an outgroup (species C). Then, the binding affinity change (deltaSVM) of the two substitutions accumulated in the red branch leading to species A was calculated on the basis of the weight list. The significance of the observed deltaSVM was evaluated by comparing it with a null distribution of deltaSVM, constructed by scoring the same number of random substitutions 10,000 times.