Figure 1

Evaluation for the different shape indices. The app. 10,000 sequences of the testing set are searched against Rfam with all 5 mentioned types of shape indices, that are namely: 1-SS_cons-, 1-consensus-, 1-hybrid-, 1-union-, and k-best-shape-index. Additionally the figure contains results for a 1-RNAalifold-shape-index (RNAalifold is used to construct a consensus structure for each family that is later transformed to a shape), a k-RNAlishapes-shape-index (the same as RNAalifold but with suboptimal structures of different shape), and the results from a complete run with the HMM filter (pink triangle in the lower right corner). 175 different parameter sets, see subsection Testing in the Evaluation, are used for the k-best-shape-index. The blue coloured data point (k _family = 3, k _query = 5, ε = 0.4) displays our recommended parameter settings as a practical trade-off between filtration ratio and sensitivity. k _family for the k-RNAlishapes-shape-index is handled as before, but the energy-index is switched off by a gigantic value for ε. The other five shape-indices use only one shape per family or per family member, so their k _family is always set to 1. Their energy-index is switched off, too.