


SDPsite is a tool for identification of protein active and other functional sites, based on spatial clustering of SDPs (specificitydetermining positions, described here) with CPs (conserved positions). The algorithm contains three parts:
Each of these parts can be run separately. Prediction of SDPsSDPs for predefined specificity groupsThe algorithm for prediction of SDPs is described in Kalinina et al. (2004) Prot Sci 13: 443456. The input data of the algorithm are a multiple protein alignment divided into specificity groups. Proteins of the same specificity group have same specificity to the ligand (or DNA, or another interacting protein), and proteins of different specificity groups may have different specificity. Let's consider each position of the alignment independently. To evaluate if the considered position is SDP, calculate its mutual information, a measure of how well the distribution of amino acids in the positions is associated with grouping bu specificity: where We introduce a number of corrections to account for properties of real biological data (discussed in Kalinina et al. (2004) Prot Sci 13: 443456). Then, using random shuffling of the column, we calculate the mean and the variance of expected column mutual information, and then a zscore of the position: A high value of Zscores indicates a position, where the amino acid distribution is much closer associated with grouping by specificity than for an average position of the alignment, and thus, which is likely to be an SDP. Given a series of Zscores corresponding to every position of the multiple alignment, one needs to evaluate the significance of the Zscores in order to tell whether the observed Zscore is sufficiently high to indicate a SDP. SDPpred uses an automated procedure for setting the thresholds based on the computation of the Bernoulli estimator. The observed Zscores are oredered by decrease: . The threshold is defined as: where Thus, we conclude that k* highestscoring position are nonrandomly distributed and designate them SDPs. Probability is called statistical significance of the set of k* positions. Automated indetification of specificity groupsSDPsite does not request userdefined specificity groups. Instead, an automated procedure for identification of specificity groups is implemented. The user is requested to provide an unrooted tree, than SDPsite performs analysis similar to Evolutionary Trace. The root of the tree is assumed to be placed in the middle of the longest path between leaves. Than a series of groupings obtained from dissection of the tree at different distance from the root is considered (see fig. below). Groups containing less than three sequences were not considered. Than for each grouping we find SDPs as described above and calculate the statistical significance P* for the obtained set of SDPs. The calculated Zscores are corrected: This correction is needed, because zscore logarithmically increases with the increase of the group size. Grouping that generates set of SDPs with the least P* is called best, or correct, grouping. Prediction of CPsThere are different ways to measure conservation score of a position. SDPsite implements one proposed be Sander and Schneider (Sander and Schneider (1991) Proteins 9: 5668): where For each C_{p}, its statistical significance is calculated. First, we calculate the background distribution, which is distribution of conservation scores of columns made up of one random position from each sequence. For each C_{p}, 10^{4} random comservation scores C_{p}^{rand} are calculated, and then zscore is computed as: As alignment of two random sequences has a nonzero weight, the obtained zscores has to be centered: Then we select the correct number of CPs using the Bernoulli estimator procedure described above. Construction of the best clusterTo construct the best cluster, SDPsite requires a 3D structure of one of the proteins of the considered alignment. On the 3D structure, it locates predicted SDPs and CPs and cluster them using layered of tightness set function algorithm (Mirkin and Muchnik (2002) Appl Math Lett 15(2): 147151). The layered clusters are constructed as follows. The graph vertices correspond to the set of SDPs and CPs on the 3D structure. Initial cluster H_{0} includes all graoh vertices. Then, for each vertex i, ins weight is calculated: where j indexes all the other vertices from H_{0} and is calculated as follows: where Then, find set of vertices , for which achieves its minimum and equals . Construct cluster . This procedure is repeated until the empty set is achieved. Thus a series of clusters is constructed: . The cluster, for which , is chosen as the most significant, or best, cluster. Format requirementsThe alignment is submitted in the standard GDE format, e.g.: The tree is submitted in the Newick format: 