Index
About CoeVizCoeViz is a web-based tool for analysis and visualization of coevolution of protein residues. The server computes pairwise coevolution scores using three metrics: Mutual Information, Chi-square Statistic, and Pearson correlation. Also, an option for computing conservation scores based on the Joint Shannon Entropy is provided. Interactive analysis includes dendrogram of clustered residues (hierarchical cluster tree), zoomable heatmap, circular diagrams of inter-residue relationships, and the mapping of residue clusters to a protein sequence or 3D structure. The tool is part of the POLYVIEW-2D protein structure visualization server and available from the resulting pages of POLYVIEW-2D. Terms of use and disclaimerAll images generated by the CoeViz server can be FREEly saved, printed, and distributed by means of any media without our written permission for academic and non-commercial purposes. However, the use of CoeViz pictures SHOULD be acknowledged by a reference to the server. The use of the CoeViz server is at your own risk and no liability is accepted for any loss or damage arising through the use of the web site and protein annotations generated by the server. References to cite the serverF.N. Baker and A. Porollo (2016) CoeViz: a web-based tool for coevolution analysis of protein residues, BMC Bioinformatics, 17: 119. MethodsImplementationUnless the MSA for a given protein is provided by the user, alignments are generated on the server side using three iterations of PSI-BLAST with the profile-inclusion threshold of expect (e)-value = 0.001 and the number of aligned sequences 2000. The sequence homology search can be done against the Pfam or NCBI NR databases. The latter database is represented by three options: full and reduced to 90 % or 70 % sequence identity by CD-HIT. Interactive analysis and visualization include: cluster tree, zoomable heat map, and relationship circular diagrams. Covariance and conservation metricsCoevolution scores are computed from the MSA using three different covariance metrics: mutual information (MI, Eq. 1), chi-square statistic (χ^{2}, Eq. 2), and Pearson correlation (r, Eq. 3). Conservation is defined by the joint Shannon entropy (S, Eq. 4). Each metric, in turn, is computed using four weighting schemes: weighted by sequence dissimilarity or sequence gapping in the alignment (Eqs. 5 and 6), by phylogeny background (Eq. 7), and non-weighted. MI scores have an additional adjustment using the average product correction (APC, Eq. 8) to produce MIp scores (Eq. 9). All metrics based on frequencies are computed using four states as possible combinations of amino acids at two positions (i and j), where each amino acid is either equal (X) or not equal (!X) to the one in the query sequence.
where x={X; !X} and y={Y; !Y}; p(s) is the observed frequency of state s={x; y; x,y}; N_{eff} is the effective sum of weights of alignments where both positions are not gaps. w_{sl} is a weighted count of state s, which is equal to 1 for non-weighted scores, 1–(percent of sequence identity) or 1–(percent of gaps) of the alignment l for weighting by sequence dissimilarity or alignment gapping, respectively, and w_{a}^{ph} for weighting by phylogeny. w_{a}^{ph} is a weight for sequence A^{a} in the MSA of N total sequences that equals to one over the number of sequences A^{b} in the MSA that have at least 80% sequence identity to A^{a}. s_{il} is a similarity score that quantifies the change of an amino acid at position i to the one in the aligned sequence l. s¯_{l} and σ_{i} are mean and standard deviation, respectively, of all similarity scores of changes for a given position represented across the all sequences aligned to the query. Similarity scores are taken from the position specific similarity matrix (PSSM) generated by PSI-BLAST. λ is a pseudo count, which is equal to 1 for all metrics here.
where MI(a,x¯) is the mean MI of column a, and MI¯¯ is the overall mean MI. Negative values of MIp scores are assigned to 0, and then all MI scores are min-max normalized to range [0, 1]. S is normalized to the same range by factor 1/log (4). χ2 values are converted to the corresponding cumulative probabilities at degree of freedom (df) = 1. Scores for each metric are organized in symmetrical matrices with the main diagonal presenting plain or weighted frequencies, as defined above, of each individual residue for MI-and χ2-based metrics, and the individual Shannon entropies using 20 states (20 amino acids) for S-based metric. Individual entropies are computed using probability part of the PSSM files from the PSI-BLAST output and normalized to range [0, 1] by factor 1/log(20). Residues of the query protein are clustered using hierarchical clustering with the complete linkage method. Prior to clustering, negative r scores are assigned to 0; MI, r, and χ2 scores are converted to distances by 1–score transformation. ManualThe CoeViz layout is split into 4 main panes: the header, tree diagram, navigation pane, and heatmap. Additional pane is shown upon request for a given cell of the heatmap. It presents inter-residue relationships using an interactive circular diagram. Detailed decription of the panes follows below.
ExampleFigure 1 illustrates how CoeViz can help identify functionally important residues using a peptidase from baker’s yeast (SwissProt ID: DUG1_YEAST) as an example. Dug1p is a Cys-Gly dipeptidase and belongs to the M20A family of metallopeptidases. The enzyme requires two Zinc ions in the active site to cleave the substrate. The structure of the enzyme is resolved (PDB ID: 4G1P), which makes it easier to map findings into 3D structure for illustration. Figure 1A shows a fragment of the clustering tree based on χ2 scores weighted by phylogeny with zoom in on the residues binding Zn (H102, D137, E172, H450) and a catalytic residue (E171) that are clustered together. Interestingly, R348 is in the same cluster. When the residues are mapped to the resolved 3D structure, where the enzyme is co-crystallized with the substrate and Zn ions in the active site, R348 appears to be on the opposite side of the active site cavity and in contact with the substrate suggesting its role in substrate recognition and positioning the dipeptide into the catalytic center (Figure 1B). A circular diagram with the closest relationships to a given residue (e.g., E171) can be an alternative way for a quick identification of covarying residues (Figure 1C).
Click on the buttons below to view interactive examples of using CoeViz to analyze baker's yeast metallopeptidase Dug1p (PDB ID: 4G1P). AcknowledgementsFor a complete list of people involed in related projects, software used, and funding, please visit the dedicated page. Last update of the document: March, 2016 Back to the POLYVIEW-2D/CoeViz server home page |