Figure 3

Workflow of a shape index construction. The example demonstrates the process of index construction for the family RF00432. For each of the five shape abstraction levels, an independent shape index has to be computed. For this purpose, every single sequence of each Rfam family (currently 574 families in Rfam 8.0) has to be abstracted with k _family shapes (k _family = 2 for the example). This is done by RNAshapes. After the shapes were computed, they are stored in the hash based shape index, where they later serve for the query look-ups. The index is a hierarchically arranged, 3-dimensional search structure – in descending search order of shape-abstraction-level, shape-string and shrep-energy. The example also displays a very characteristic phenomenon: the weaker an abstraction is, the greater is the variance of shapes and the smaller are the family-lists that are associated with these shapes. The last fact is not shown directly, but can be recognized by the increasing energy-interval-size.