Flowchart summarizing the major steps in the methodology followed throughout this analysis: A phylogenetic analysis using both whole-genome sequence (if applicable) and the amino acid sequence of the core gene products was carried out enabling the construction of the reference tree topology for each genus. In a second step, a comparative analysis (genomewise) was performed between the chromosomes of each genus and the corresponding outgroups, leading to the identification of regions with limited phylogenetic distribution. In a third step, a maximum parsimony model (based on the reference tree topology) was applied in order to differentiate gene gain from gene loss events and exclude regions with limited phylogenetic distribution due to a gene loss event. The remaining regions formed the positive control data set (i.e., putative GIs) of this analysis. The negative control data set (i.e., non-GIs) was built implementing a random sampling approach, sampling regions only within the inter-GI parts of the chromosome; both positive and negative examples were annotated structurally. In a final step, the structural features of each region were used as input vectors to a machine learning method (RVM) leading to the construction of structural GI models.
