Figure 2

Feature selections. Diagram representing the different types of features used by our predictor. For the local density of functional elements, the categories are coding amniote, non-coding amniote, coding metatheria, non-coding methateria within a 2 Mb window. For each of these categories, the distance from the candidate breakpoint to the center of inter-marker region considered is taken in account and various weighting factors termed α are applied. The effect of an α factor of 2 is presented at the top of the figure. In that case markers close to the breakpoint have a high weight whereas distant markers only bring a small contribution. The functional association between regulatory regions and genes is a different category of features. Each non-coding element is associated with a set of genes depending on the value of the β factor, which determines how far the association between non-coding and coding regions will be considered (β = 2 is illustrated). For a given inter-marker region, the predictor will test if the region overlaps such associations. In total, 29 features are considered.