Fig. 2.

Schematic representation of KMAD’s scoring function; i and j are positions in the profile and in the sequence, respectively; profile_i is a vector of length 20 that contains the frequencies of the 20 amino acid types m at position i in the alignment that resulted from the previous iteration (in the first iteration all profile values are 0 or 1, according to the first sequence). D is an IDP-specific 20 × 20 substitution matrix (Midic et al., 2009). The residue score is the convolution of the profile vector for i and the D column for j. ${\omega }_{f_t}$ are weights for the three feature types f_t (${\omega }_{f_t}$ weights optionally can be set by the user). $U_{f,i}$ values for PTMs and SLiMs are determined from the conservation of that feature at i. $U_{f,i}$ values for domains are the conservation of the domain at position i minus the conservation of all other domains at that position. The terms R relate to the perceived reliability (PR) of the feature; experimentally determined PTMs, for example, weigh three times higher than predicted ones. $R_{f,i,j}$ is the product of the average PR of the feature f at position i in the profile and the PR of this feature at position j in the sequence. The metascore for j is the weighted sum of scores for all observed features