Figure 8.

Relation between prediction accuracy and the number of variables included in the model. The fractions of variables included in (A) the internal motion prediction model and (B) the external_short motion prediction model. The histogram shows fractions of each variable category in the model when the variables are divided into four categories according to the feature type (see the Methods section). The plot shows percentages of the number of variables included in a downsized model against the number of variables included in the original model. The fraction of the variable category and percentage of the number of variables are respectively represented on the left and right sides' vertical axes. Although all variables were computed from the original model, the center of the secondary structure (CS), periphery of the secondary structure (PS), and remote area from the secondary structure (RS) were computed from downsized models. The CS, PS, and RS respectively signify categories which classify an amino acid according to the secondary structure (see the Methods section). Furthermore, S, A, P, and M respectively signify the names of feature groups described in the Methods section, and corresponding to secondary structure, ASA, physicochem., and mobility. The value of the Gini index was set to 1.7 in (A) and 0.7 in (B). The external_long prediction model showed similar tendencies to those of the external_short prediction model (data not shown). These results were obtained using the method that implemented psipred and sable.