Protein Structure and Function: Application of Bioinformatics Methods - John Rigden 2014

Spatial Motifs
Overview of Methods
Interpretation of Results

A wide variety of web servers are available for comparing a Cell/13.html">Protein Structure of interest with Databases of Structural motifs (Table 8.1). What Conclusions can be drawn about a protein's function if a specific motif is identified within its structure? Determining whether a match is statistically significant, and if so, what biological implications it carries, requires addressing several key questions.

The putative function largely depends on how the motif was derived. If the motif consists of a set of residues experimentally proven to be essential for a specific function, identifying it in a query protein suggests that the protein may share this function. Similarly, discovering a motif extracted from the structures of multiple Proteins with a similar function (positive Examples) implies that the query protein likely performs the same function, especially if the motif was designed to exclude negative examples. However, the exact function assigned to a structure depends heavily on the criteria used to define the positive examples. For instance, if all positive examples bind adenosine yet catalyze diverse reactions, finding their common motif in a query structure implies adenosine binding, but not necessarily adenosine deaminase activity. If the set of positive examples comprises members of a single SCOP superfamily, identifying the motif in the query protein indicates that it belongs to that same superfamily, though more specific functional predictions cannot be made.

Ideally, the set of positive examples used for motif discovery should be as diverse as possible while sharing a common property, whereas negative examples should closely resemble the positive ones but lack that specific property. In practice, training sets of positive and negative examples are rarely ideal, and some or all derived structural motifs may ultimately reflect common evolutionary origins or coincidental similarities rather than a shared function.

Interpreting motifs extracted from single structures is inherently subjective. As with sequence analysis, the query structure is annotated by analogy to the protein that served as the source of the best-matching motif. While identifying a motif corresponding to a binding site is traditionally interpreted as a prediction of binding Specificity, it is also frequently used (with varying degrees of accuracy) to infer other annotations, such as catalytic activity or membership in a specific family or superfamily. It should be noted that mere spatial proximity to a Ligand does not guarantee that a residue is crucial for binding or catalysis, as it may have arisen there through mutation. Apart from utilizing residues surrounding a ligand, another way to extract a motif from a single structure is to rely on annotations provided in the PDB file (SITE records). Although these records are intended to list residues forming binding sites, the Biological Significance of such motifs remains ambiguous due to a lack of strict criteria for which residues should be included. Furthermore, many PDB files of ligand-binding structures contain no SITE records at all.

Class="center">Table 8.1. Web servers for searching and comparing structural motifs

^a Sourced from publications, web services, or author communications; links may be outdated.

^b For non-commercial use only.

Another factor worth considering is the stringency of motif matching, which depends on motif size and representation, as well as the quantitative criteria and numerical thresholds applied during the search. Motif representations based solely on alpha-carbon atoms are less specific than those incorporating side-chain atoms. Structural motifs spanning only a small number of residues are similarly less specific. Strict matching parameters (such as restricting matches to identical residue types or enforcing a low RMSD threshold) may restrict results to closely related homologs, even though relaxed criteria could yield meaningful matches with more evolutionarily distant proteins.

Most Methods employ scoring functions to rank hits and evaluate match quality. For instance, RMSD values indicate the degree of geometric proximity between structural motif points and corresponding atoms in the target structure. While RMSD is a suitable metric for ranking motifs by their similarity to a reference template, it cannot be reliably used to compare motifs of different sizes. Furthermore, certain motifs are inherently more prone to accidental matches simply because they contain a higher proportion of common residues. To address these issues and improve hit ranking, some methods calculate statistical significance (P-value) or expectation values (E-value). However, their applicability comes with certain limitations, as these values depend heavily on foundational assumptions within the statistical model and the data used for its parameterization.

Regardless of how a structural motif was derived, it can be tested against a benchmark set of structures. When this benchmark includes reliable positive and negative examples, performance can be evaluated in terms of sensitivity (The ability to correctly retrieve positive examples) and specificity (the ability to exclude negative examples). When the benchmark consists solely of negative examples, the resulting RMSD distribution can be used to estimate the statistical significance of matches to that motif. The reliability of these derived metrics depends on having a sufficiently large and representative test dataset.

A consensus approach can be highly beneficial, where multiple hits to related motifs or consistent results obtained from different software tools and databases converge on a unified prediction.

Finally, standard biological common sense should always be applied. For example, a high-scoring match to an active-site motif is implausible if the protein lacks a corresponding substrate-binding pocket. Moreover, statistical significance and biological relevance are not synonymous; a biologically significant motif may fail to achieve high statistical significance compared to a structurally similar but functionally irrelevant motif. Therefore, before using motifs to infer the function or other characteristics of a protein, it is crucial to visually inspect the matches and evaluate them against sound biological criteria.