The genome-wide identification of pairs of interacting proteins is an important step in the elucidation of cell regulatory mechanisms1,2. Much of our current knowledge derives from high-throughput techniques such as yeast two hybrid and affinity purification3, as well as from manual curation of experiments on individual systems4. A variety of computational approaches based, for example, on sequence homology, gene co-expression, and phylogenetic profiles have also been developed for the genome-wide inference of protein-protein interactions (PPIs)5,6. Yet, comparative studies suggest that the development of accurate and complete repertoires of PPIs is still in its early stages7–9. Here we show that three-dimensional structural information can be used to predict PPIs with an accuracy and coverage that are superior to predictions based on non-structural evidence. Moreover, an algorithm, PrePPI, that combines structural information with other functional clues is comparable in accuracy to high-throughput experiments, yielding over 30,000 high confidence interactions for yeast and over 300,000 for human. Experimental tests of a number of predictions demonstrate the ability of the PrePPI algorithm to identify unexpected PPIs of significant biological interest. The surprising effectiveness of three-dimensional structural information can be attributed to the use of homology models combined with the exploitation of both close and remote geometric relationships between proteins.

Interfaces between organelles are emerging as critical platforms for many biological responses in eukaryotic cells. In yeast, the ERMES complex is an endoplasmic reticulum (ER)–mitochondria tether composed of four proteins, three of which contain a SMP (synaptotagmin-like mitochondrial-lipid binding protein) domain. No functional ortholog for any ERMES protein has been identified in metazoans. Here, we identified PDZD8 as an ER protein present at ER-mitochondria contacts. The SMP domain of PDZD8 is functionally orthologous to the SMP domain found in yeast Mmm1. PDZD8 was necessary for the formation of ER-mitochondria contacts in mammalian cells. In neurons, PDZD8 was required for calcium ion (Ca2+) uptake by mitochondria after synaptically induced Ca2+-release from ER and thereby regulated cytoplasmic Ca2+ dynamics. Thus, PDZD8 represents a critical ER-mitochondria tethering protein in metazoans. We suggest that ER-mitochondria coupling is involved in the regulation of dendritic Ca2+ dynamics in mammalian neurons.

We participated in the fold recognition and homology sections of CASP5 using primarily in-house software. The central feature of our structure prediction strategy involved the ability to generate good sequence-to-structure alignments and to quickly transform them into models that could be evaluated both with energy-based methods and manually. The in-house tools we used include: a) HMAP (Hybrid Multidimensional Alignment Profile)-a profile-to-profile alignment method that is derived from sequence-enhanced multiple structure alignments in core regions, and sequence motifs in non-structurally conserved regions. b) NEST-a fast model building program that applies an "artificial evolution" algorithm to construct a model from a given template and alignment. c) GRASP2-a new structure and alignment visualization program incorporating multiple structure superposition and domain database scanning modules. These methods were combined with model evaluation based on all atom and simplified physical-chemical energy functions. All of these methods were under development during CASP5 and consequently a great deal of manual analysis was carried out at each stage of the prediction process. This interactive model building procedure has several advantages and suggests important ways in which our and other methods can be improved, examples of which are provided.

PrePPI (http://bhapp.c2b2.columbia.edu/PrePPI) is a database that combines predicted and experimentally determined protein–protein interactions (PPIs) using a Bayesian framework. Predicted interactions are assigned probabilities of being correct, which are derived from calculated likelihood ratios (LRs) by combining structural, functional, evolutionary and expression information, with the most important contribution coming from structure. Experimentally determined interactions are compiled from a set of public databases that manually collect PPIs from the literature and are also assigned LRs. A final probability is then assigned to every interaction by combining the LRs for both predicted and experimentally determined interactions. The current version of PrePPI contains ∼2 million PPIs that have a probability more than ∼0.1 of which ∼60 000 PPIs for yeast and ∼370 000 PPIs for human are considered high confidence (probability > 0.5). The PrePPI database constitutes an integrated resource that enables users to examine aggregate information on PPIs, including both known and potentially novel interactions, and that provides structural models for many of the PPIs.

No abstract

With the advent of Systems Biology, the prediction of whether two proteins form a complex has become a problem of increased importance. A variety of experimental techniques have been applied to the problem, but three-dimensional structural information has not been widely exploited. Here we explore the range of applicability of such information by analyzing the extent to which the location of binding sites on protein surfaces is conserved among structural neighbors. We find, as expected, that interface conservation is most significant among proteins that have a clear evolutionary relationship, but that there is a significant level of conservation even among remote structural neighbors. This finding is consistent with recent evidence that information available from structural neighbors, independent of classification, should be exploited in the search for functional insights. The value of such structural information is highlighted through the development of a new protein interface prediction method, PredUs, that identifies what residues on protein surfaces are likely to participate in complexes with other proteins. The performance of PredUs, as measured through comparisons with other methods, suggests that relationships across protein structure space can be successfully exploited in the prediction of protein-protein interactions.T he knowledge of whether two proteins form a complex is a problem of central importance in the description of cellular networks and in a large number of other biological applications. Much effort has been devoted recently to high-throughput experimental determination and literature curation of protein-protein interactions (see refs. 1 and 2 for a review), and the results have been deposited into numerous databases (3, 4). In addition, a variety of computational approaches have been developed to predict protein interaction partners (2, 5-7). Three-dimensional structural information has not been widely used in large-scale studies, in part because the number of complexes for which such information is available is far smaller than the number of interactions that can be inferred by other techniques.A number of groups have shown that the use of homologous relationships can expand the range of structural information by providing plausible models for a protein complex that can then be evaluated with other methods (8-10). However, the extent to which a known 3D structure of a complex can be used reliably as a template for a model of two related proteins is unclear, especially if the relevant sequence and/or structural relationship is remote. Model reliability should, in general, increase if the proteins involved are closely related, but the use of close homologs necessarily limits the number of possible interactions that can be detected. We have recently shown (11) that the use of remote structural relationships can detect functional relationships between proteins that are obscured by classification schemes. One of the aims of the current paper is to evaluate whether structural relationships that c...

Here we carry out an examination of shape complementarity as a criterion in protein-protein docking and binding. Specifically, we examine the quality of shape complementarity as a critical determinant not only in the docking of 26 protein-protein ''bound'' complexed cases, but in particular, of 19 ''unbound'' protein-protein cases, where the structures have been determined separately. In all cases, entire molecular surfaces are utilized in the docking, with no consideration of the location of the active site, or of particular residues/ atoms in either the receptor or the ligand that participate in the binding. To evaluate the goodness of the strictly geometry-based shape complementar-ity in the docking process as compared to the main favorable and unfavorable energy components, we study systematically a potential correlation between each of these components and the root mean square deviation (RMSD) of the ''unbound'' protein-protein cases. Specifically, we examine the non-polar buried surface area, polar buried surface area, buried surface area relating to groups bearing unsatisfied buried charges, and the number of hydrogen bonds in all docked protein-protein interfaces. For these cases, where the two proteins have been crystallized separately, and where entire molecular surfaces are considered without a predefi-nition of the binding site, no correlation is observed. None of these parameters appears to consistently improve on shape complementarity in the docking of unbound molecules. These findings argue that simplicity in the docking process, utilizing geometrical shape criteria may capture many of the essential features in protein-protein docking. In particular, they further reinforce the long held notion of the importance of molecular surface shape complementarity in the binding, and hence in docking. This is particularly interesting in light of the fact that the structures of the docked pairs have been determined separately, allowing side chains on the surface of the proteins to move relatively freely. This study has been enabled by our efficient, computer vision-based docking algorithms. The fast CPU matching times, on the order of minutes on a PC, allow such large-scale docking experiments of large molecules, which may not be feasible by other techniques. Proteins 1999;36:307-317. 1999 Wiley-Liss, Inc.

Very few methods address the problem of predicting beta-barrel membrane proteins directly from sequence. One reason is that only very few high-resolution structures for transmembrane beta-barrel (TMB) proteins have been determined thus far. Here we introduced the design, statistics and results of a novel profile-based hidden Markov model for the prediction and discrimination of TMBs. The method carefully attempts to avoid over-fitting the sparse experimental data. While our model training and scoring procedures were very similar to a recently published work, the architecture and structure-based labelling were significantly different. In particular, we introduced a new definition of beta- hairpin motifs, explicit state modelling of transmembrane strands, and a log-odds whole-protein discrimination score. The resulting method reached an overall four-state (up-, down-strand, periplasmic-, outer-loop) accuracy as high as 86%. Furthermore, accurately discriminated TMB from non-TMB proteins (45% coverage at 100% accuracy). This high precision enabled the application to 72 entirely sequenced Gram-negative bacteria. We found over 164 previously uncharacterized TMB proteins at high confidence. Database searches did not implicate any of these proteins with membranes. We challenge that the vast majority of our 164 predictions will eventually be verified experimentally. All proteome predictions and the PROFtmb prediction method are available at http://www.rostlab.org/ services/PROFtmb/.