Search

The interpretation of human genetic variation is one of the greatest challenges of modern genetics. New approaches are urgently needed to prioritize variants, especially those that are rare or lack a definitive clinical interpretation. We examined 10,136,597 human missense genetic variants from GnomAD, ClinVar and UniProt. We were able to perform large-scale atom-based mapping and phenotype interpretation of 3,960,015 of these variants onto 18,874 experimental and 84,818 in house predicted three-dimensional coordinates of the human proteome. We demonstrate that 14% of amino acid substitutions from the GnomAD database that could be structurally analysed are predicted to affect protein structure (n = 568,548, of which 566,439 rare or extremely rare) and may, therefore, have a yet unknown disease-causing effect. The same is true for 19.0% (n = 6266) of variants of unknown clinical significance or conflicting interpretation reported in the ClinVar database. The results of the structural analysis are available in the dedicated web catalogue Missense3D-DB (http://missense3d.bc.ic.ac.uk/). For each of the 4 M variants, the results of the structural analysis are presented in a friendly concise format that can be included in clinical genetic reports. A detailed report of the structural analysis is also available for the non-experts in structural biology. Population frequency and predictions from SIFT and PolyPhen are included for a more comprehensive variant interpretation. This is the first large-scale atom-based structural interpretation of human genetic variation and offers geneticists and the biomedical community a new approach to genetic variant interpretation.

A big challenge in current systems biology research arises when different types of data must be accessed from separate sources and visualized using separate tools. The high cognitive load required to navigate such a workflow is detrimental to hypothesis generation. Accordingly, there is a need for a robust research platform that incorporates all data and provides integrated search, analysis, and visualization features through a single portal. Here, we present ePlant (http://bar.utoronto.ca/eplant), a visual analytic tool for exploring multiple levels of Arabidopsis thaliana data through a zoomable user interface. ePlant connects to several publicly available web services to download genome, proteome, interactome, transcriptome, and 3D molecular structure data for one or more genes or gene products of interest. Data are displayed with a set of visualization tools that are presented using a conceptual hierarchy from big to small, and many of the tools combine information from more than one data type. We describe the development of ePlant in this article and present several examples illustrating its integrative features for hypothesis generation. We also describe the process of deploying ePlant as an “app” on Araport. Building on readily available web services, the code for ePlant is freely available for any other biological species research.

Pleiotropy is the phenomenon by which the same gene can result in multiple phenotypes. Pleiotropic proteins are emerging as important contributors to rare and common disorders. Nevertheless, little is known on the mechanisms underlying pleiotropy and the characteristic of pleiotropic proteins. We analyzed disease-causing proteins reported in UniProt and observed that 12% are pleiotropic (variants in the same protein cause more than one disease). Pleiotropic proteins were enriched in deleterious and rare variants, but not in common variants. Pleiotropic proteins were more likely to be involved in the pathogenesis of neoplasms, neurological, and circulatory diseases and congenital malformations, whereas non-pleiotropic proteins in endocrine and metabolic disorders. Pleiotropic proteins were more essential and had a higher number of interacting partners compared with non-pleiotropic proteins. Significantly more pleiotropic than non-pleiotropic proteins contained at least one intrinsically long disordered region (P

Non-synonymous single nucleotide polymorphisms (nsSNPs) are single base changes leading to a change to the amino acid sequence of the encoded protein. Many of these variants are associated with disease, so nsSNPs have been well studied, with studies looking at the effects of nsSNPs on individual proteins, for example, on stability and enzyme active sites. In recent years, the impact of nsSNPs upon protein-protein interactions has also been investigated, giving a greater insight into the mechanisms by which nsSNPs can lead to disease. In this review, we summarize these studies, looking at the various mechanisms by which nsSNPs can affect protein-protein interactions. We focus on structural changes that can impair interaction, changes to disorder, gain of interaction, and post-translational modifications before looking at some examples of nsSNPs at human-pathogen protein-protein interfaces and the analysis of nsSNPs from a network perspective.

The poly-l-proline type II (PPII) helix in recent years has emerged clearly as a structural class not only of fibrillar proteins (in collagen, PPII is a dominant conformation) but also of the folded and unfolded proteins. Although much less abundant in folded proteins than the α-helix and β-structure, the left-handed, extended PPII helix represents the only frequently occurring regular structure apart from these two structure classes. Natively unfolded proteins have a high content of the PPII helices identified by spectroscopic methods. Apart from the structural function, PPII is favorable for protein-protein and protein-nucleic acid interactions and plays a major role in signal transduction and protein complex assembly, as this structure is often found in binding sites, specifically binding sites of widely spread SH3 domains. PPII helices do not necessarily contain proline, but proline has high PPII propensity. Commonly occurring proline-rich regions, serving as recognition sites, are likely to have PPII structure. PPII helices are involved in transcription, cell motility, self-assembly, elasticity, and bacterial and viral pathogenesis, and has an important structural role in amyloidogenic proteins. However, PPII helices are not always assigned in experimentally solved structures, and they are rarely used in protein structure modeling. We aim to give an overview of this structural class and of the place it holds in our current understanding of protein structure and function. This review is subdivided into three main parts: the first part covers PPII helices in unfolded peptides and proteins, the second part includes studies of the PPII helices in folded proteins, and the third part discusses the functional role of the PPII.

The acid-labile subunit (ALS) is the main regulator of IGF1 and IGF2 bioavailability. ALS deficiency caused by mutations in the ALS (IGFALS) gene often results in mild short stature in adulthood. Little is known about the ALS structure–function relationship. A structural model built in 1999 suggested a doughnut shape, which has never been observed in the leucine-rich repeat (LRR) superfamily, to which ALS belongs. In this study, we built a new ALS structural model, analysed its glycosylation and charge distribution and studied mechanisms by which missense mutations affect protein structure. We used three structure prediction servers and integrated their results with information derived from ALS experimental studies. The ALS model was built at high confidence using Toll-like receptor protein templates and resembled a horseshoe with an extensively negatively charged concave surface. Enrichment in prolines and disulphide bonds was found at the ALS N- and C-termini. Moreover, seven N-glycosylation sites were identified and mapped. ALS mutations were predicted to affect protein structure by causing loss of hydrophobic interactions (p.Leu134Gln), alteration of the amino acid backbone (p.Leu241Pro, p.Leu172Phe and p.Leu244Phe), loss of disulphide bridges (p.Cys60Ser and p.Cys540Arg), change in structural constrains (p.Pro73Leu), creation of novel glycosylation sites (p.Asp440Asn) or alteration of LRRs (p.Asn276Ser). In conclusion, our ALS structural model was identified as a highly confident prediction by three independent methods and disagrees with the previously published ALS model. The new model allowed us to analyse the ALS core and its caps and to interpret the potential structural effects of ALS mutations.

Motivation: Analysis of protein–protein interaction networks (PPINs) at the system level has become increasingly important in understanding biological processes. Comparison of the interactomes of different species not only provides a better understanding of species evolution but also helps with detecting conserved functional components and in function prediction. Method and Results: Here we report a PPIN alignment method, called PINALOG, which combines information from protein sequence, function and network topology. Alignment of human and yeast PPINs reveals several conserved subnetworks between them that participate in similar biological processes, notably the proteasome and transcription related processes. PINALOG has been tested for its power in protein complex prediction as well as function prediction. Comparison with PSI-BLAST in predicting protein function in the twilight zone also shows that PINALOG is valuable in predicting protein function. Availability and implementation: The PINALOG web-server is freely available from http://www.sbg.bio.ic.ac.uk/~pinalog. The PINALOG program and associated data are available from the Download section of the web-server. Contact: m.sternberg@imperial.ac.uk Supplementary information: Supplementary data are available at Bioinformatics online.

Motivation: Databases of sequenced genomes are widely used to characterize the structure, function and evolutionary relationships of proteins. The ability to discern such relationships is widely expected to grow as sequencing projects provide novel information, bridging gaps in our map of the protein universe. Results: We have plotted our progress in protein sequencing over the last two decades and found that the rate of novel sequence discovery is in a sustained period of decline. Consequently, PSI-BLAST, the most widely used method to detect remote evolutionary relationships, which relies upon the accumulation of novel sequence data, is now showing a plateau in performance. We interpret this trend as signalling our approach to a representative map of the protein universe and discuss its implications. Contact: daniel.chubb01@imperial.ac.uk Supplementary information: Supplementary data are available at Bioinformatics online.

Charles Darwin's theory of evolution was based on studies of biology at the species level. In the time since his death, studies at the molecular level have confirmed his ideas about the kinship of all life on Earth and have provided a wealth of detail about the evolutionary relationships between different species and a deeper understanding of the finer workings of natural selection. We now have a wealth of data, including the genome sequences of a wide range of organisms, an even larger number of protein sequences, a significant knowledge of the three-dimensional structures of proteins, DNA and other biological molecules, and a huge body of information about the operation of these molecules as systems in the molecular machinery of all living things. This issue of Biochemical Society Transactions contains papers from oral presentations given at a Biochemical Society Focused Meeting to commemorate the 200th Anniversary of Charles Darwin's birth, held on 26–27 January 2009 at the Wellcome Trust Conference Centre, Cambridge. The talks reported on some of the insights into evolution which have been obtained from the study of protein sequences, structures and systems.

The critical assessment of protein structure prediction experiment is a blind assessment of the prediction of protein structure and related topics including function prediction. We present our results in the function/binding site prediction category. Our approach to identify binding sites combined the use of the predicted structure of the targets with both residue conservation and the location of ligands bound to homologous structures. We obtained an average coverage of 83% and 56% accuracy. Analysis of our predictions suggests that over-prediction reduces the accuracy obtained due to large areas of conservation around the binding site that do not bind the ligand. In some proteins such conserved residues may have a functional role. A server version of our method will soon be available.

In chemoinformatics, searching for compounds which are structurally diverse and share a biological activity is called scaffold hopping. Scaffold hopping is important since it can be used to obtain alternative structures when the compound under development has unexpected side-effects. Pharmaceutical companies use scaffold hopping when they wish to circumvent prior patents for targets of interest. We propose a new method for scaffold hopping using inductive logic programming (ILP). ILP uses the observed spatial relationships between pharmacophore types in pretested active and inactive compounds and learns human-readable rules describing the diverse structures of active compounds. The ILP-based scaffold hopping method is compared to two previous algorithms (chemically advanced template search, CATS, and CATS3D) on 10 data sets with diverse scaffolds. The comparison shows that the ILP-based method is significantly better than random selection while the other two algorithms are not. In addition, the ILP-based method retrieves new active scaffolds which were not found by CATS and CATS3D. The results show that the ILP-based method is at least as good as the other methods in this study. ILP produces human-readable rules, which makes it possible to identify the three-dimensional features that lead to scaffold hopping. A minor variant of a rule learnt by ILP for scaffold hopping was subsequently found to cover an inhibitor identified by an independent study. This provides a successful result in a blind trial of the effectiveness of ILP to generate rules for scaffold hopping. We conclude that ILP provides a valuable new approach for scaffold hopping.

Motivation: Reliable structural modelling of protein–protein complexes has widespread application, from drug design to advancing our knowledge of protein interactions and function. This work addresses three important issues in protein–protein docking: implementing backbone flexibility, incorporating prior indications from experiment and bioinformatics, and providing public access via a server. 3D-Garden (Global And Restrained Docking Exploration Nexus), our benchmarked and server-ready flexible docking system, allows sophisticated programming of surface patches by the user via a facet representation of the interactors’ molecular surfaces (generated with the marching cubes algorithm). Flexibility is implemented as a weighted exhaustive conformer search for each clashing pair of molecular branches in a set of 5000 models filtered from around ∼340 000 initially. Results: In a non-global assessment, carried out strictly according to the protocols for number of models considered and model quality of the Critical Assessment of Protein Interactions (CAPRI) experiment, over the widely-used Benchmark 2.0 of 84 complexes, 3D-Garden identifies a set of ten models containing an acceptable or better model in 29/45 test cases, including one with large conformational change. In 19/45 cases an acceptable or better model is ranked first or second out of 340 000 candidates. Availability: http://www.sbg.bio.ic.ac.uk/3dgarden (server) Contact: v.lesk@ic.ac.uk Supplementary information: Supplementary data are available at Bioinformatics online.

Despite the increased recent use of protein-ligand and protein-protein docking in the drug discovery process due to the increases in computational power, the difficulty of accurately ranking the binding affinities of a series of ligands or a series of proteins docked to a protein receptor remains largely unsolved. This problem is of major concern in lead optimization procedures and has lead to the development of scoring functions tailored to rank the binding affinities of a series of ligands to a specific system. However, such methods can take a long time to develop and their transferability to other systems remains open to question. Here we demonstrate that given a suitable amount of background information a new approach using support vector inductive logic programming (SVILP) can be used to produce system-specific scoring functions. Inductive logic programming (ILP) learns logic-based rules for a given dataset that can be used to describe properties of each member of the set in a qualitative manner. By combining ILP with support vector machine regression, a quantitative set of rules can be obtained. SVILP has previously been used in a biological context to examine datasets containing a series of singular molecular structures and properties. Here we describe the use of SVILP to produce binding affinity predictions of a series of ligands to a particular protein. We also for the first time examine the applicability of SVILP techniques to datasets consisting of protein-ligand complexes. Our results show that SVILP performs comparably with other state-of-the-art methods on five protein-ligand systems as judged by similar cross-validated squares of their correlation coefficients. A McNemar test comparing SVILP to CoMFA and CoMSIA across the five systems indicates our method to be significantly better on one occasion. The ability to graphically display and understand the SVILP-produced rules is demonstrated and this feature of ILP can be used to derive hypothesis for future ligand design in lead optimization procedures. The approach can readily be extended to evaluate the binding affinities of a series of protein-protein complexes.

The increasing interest in systems biology has resulted in extensive experimental data describing networks of interactions (or associations) between molecules in metabolism, protein-protein interactions and gene regulation. Comparative analysis of these networks is central to understanding biological systems. We report a novel method (PHUNKEE: Pairing subgrapHs Using NetworK Environment Equivalence) by which similar subgraphs in a pair of networks can be identified. Like other methods, PHUNKEE explicitly considers the graphical form of the data and allows for gaps. However, it is novel in that it includes information about the context of the subgraph within the adjacent network. We also explore a new approach to quantifying the statistical significance of matching subgraphs. We report similar subgraphs in metabolic pathways and in protein-protein interaction networks. The most similar metabolic subgraphs were generally found to occur in processes central to all life, such as purine, pyrimidine and amino acid metabolism. The most similar pairs of subgraphs found in the protein-protein interaction networks of Drosophila melanogaster and Saccharomyces cerevisiae also include central processes such as cell division but, interestingly, also include protein sub-networks involved in pre-mRNA processing. The inclusion of network context information in the comparison of protein interaction networks increased the number of similar subgraphs found consisting of proteins involved in the same functional process. This could have implications for the prediction of protein function.

Despite being hugely important in biological processes, allostery is poorly understood and no universal mechanism has been discovered. Allosteric drugs are a largely unexplored prospect with many potential advantages over orthosteric drugs. Computational methods to predict allosteric sites on proteins are needed to aid the discovery of allosteric drugs, as well as to advance our fundamental understanding of allostery. AlloPred, a novel method to predict allosteric pockets on proteins, was developed. AlloPred uses perturbation of normal modes alongside pocket descriptors in a machine learning approach that ranks the pockets on a protein. AlloPred ranked an allosteric pocket top for 23 out of 40 known allosteric proteins, showing comparable and complementary performance to two existing methods. In 28 of 40 cases an allosteric pocket was ranked first or second. The AlloPred web server, freely available at http://www.sbg.bio.ic.ac.uk/allopred/home , allows visualisation and analysis of predictions. The source code and dataset information are also available from this site. Perturbation of normal modes can enhance our ability to predict allosteric sites on proteins. Computational methods such as AlloPred assist drug discovery efforts by suggesting sites on proteins for further experimental study.

We investigate the extent to which the conformational fluctuations of proteins in solution reflect the conformational changes that they undergo when they form binary protein-protein complexes. To do this, we study a set of 41 proteins that form such complexes and whose three-dimensional structures are known, both bound in the complex and unbound. We carry out molecular dynamics simulations of each protein, starting from the unbound structure, and analyze the resulting conformational fluctuations in trajectories of 5 ns in length, comparing with the structure in the complex. It is found that fluctuations take some parts of the molecules into regions of conformational space close to the bound state (or give information about it), but at no point in the simulation does each protein as whole sample the complete bound state. Subsequent use of conformations from a clustered MD ensemble in rigid-body docking is nevertheless partially successful when compared to docking the unbound conformations, as long as the unbound conformations are themselves included with the MD conformations and the whole globally rescored. For one key example where sub-domain motion is present, a ribonuclease inhibitor, principal components analysis of the MD was applied and was also able to produce conformations for docking that gave enhanced results compared to the unbound. The most significant finding is that core interface residues show a tendency to be less mobile (by size of fluctuation or entropy) than the rest of the surface even when the other binding partner is absent, and conversely the peripheral interface residues are more mobile. This surprising result, consistent across up to 40 of the 41 proteins, suggests different roles for these regions in protein recognition and binding, and suggests ways that docking algorithms could be improved by treating these regions differently in the docking process.

Here, we used data of complete genomes to study comparatively the metabolism of different species. We built phenetic trees based on the enzymatic functions present in different parts of metabolism. Seven broad metabolic classes, comprising a total of 69 metabolic pathways, were comparatively analyzed for 27 fully sequenced organisms of the domains Eukarya, Bacteria and Archaea. Phylogenetic profiles based on the presence/absence of enzymatic functions for each metabolic class were determined and distance matrices for all the organisms were then derived from the profiles. Unrooted phenetic trees based upon the matrices revealed the distribution of the organisms according to their metabolic capabilities, reflecting the ecological pressures and adaptations that those species underwent during their evolution. We found that organisms that are closely related in phylogenetic terms could be distantly related metabolically and that the opposite is also true. For example, obligate bacterial pathogens were usually grouped together in our metabolic trees, demonstrating that obligate pathogens share common metabolic features regardless of their diverse phylogenetic origins. The branching order of proteobacteria often did not match their classical phylogenetic classification and Gram-positive bacteria showed diverse metabolic affinities. Archaea were found to be metabolically as distant from free-living bacteria as from eukaryotes, and sometimes were placed close to the metabolically highly specialized group of obligate bacterial pathogens. Metabolic trees represent an integrative approach for the comparison of the evolution of the metabolism and its correlation with the evolution of the genome, helping to find new relationships in the tree of life.

We present a systematic study of the clustering of genes within the human genome based on homology inferred from both sequence and structural similarity. The 3D-Genomics automated proteome annotation pipeline (www.sbg.bio.ic.ac.uk/3dgenomics) was utilised to infer homology for each protein domain in the genome, for the 26 superfamilies most highly represented in the Structural Classification Of Proteins (SCOP) database. This approach enabled us to identify homologues that could not be detected by sequence-based methods alone. For each superfamily, we investigated the distribution, both within and among chromosomes, of genes encoding at least one domain within the superfamily. The results indicate a diversity of clustering behaviours: some superfamilies showed no evidence of any clustering, and others displayed significant clustering either within or among chromosomes, or both. Removal of tandem repeats reduced the levels of clustering observed, but some superfamilies still displayed highly significant clustering. Thus, our study suggests that either the process of gene duplication, or the evolution of the resulting clusters, differs between structural superfamilies.

The 3D-GENOMICS database (http://www.sbg.bio. ic.ac.uk/3dgenomics/) provides structural annotations for proteins from sequenced genomes. In August 2003 the database included data for 93 proteomes. The annotations stored in the database include homologous sequences from various sequence databases, domains from SCOP and Pfam, patterns from Prosite and other predicted sequence features such as transmembrane regions and coiled coils. In addition to annotations at the sequence level, several precomputed cross- proteome comparative analyses are available based on SCOP domain superfamily composition. Annotations are available to the user via a web interface to the database. Multiple points of entry are available so that a user is able to: (i) directly access annotations for a single protein sequence via keywords or accession codes, (ii) examine a sequence of interest chosen from a summary of annotations for a particular proteome, or (iii) access precomputed frequency-based cross-proteome comparative analyses.

The application of Inductive Logic Programming (ILP), a form of machine learning, to derive structure activity relationships (SAR) and to discover pharmacophores is reported. The ILP approach was initially applied to model 1D SARs in terms of the attributes of the molecules. Subsequently 2D ILP SARs were developed describing chemical connectivity. Finally ILP has been used to model 3D SARs in which the conformation of the pharmacophore can be described. ILP has advantages over many other widely used methods as it can reason with relations and hence discover chemical substructures and 3D features without these aspects having been explicitly encoded prior to learning. In particular, there is no requirement for a structural superposition. Additionally, the results of ILP provide chemical descriptions that can readily be understood by a medicinal chemist. In several trials, ILP-based SARs have been shown to be significantly more accurate than widely-used methods.

CAPRI is a communitywide experiment to assess the capacity of protein-docking methods to predict protein-protein interactions. Nineteen groups participated in rounds 1 and 2 of CAPRI and submitted blind structure predictions for seven protein-protein complexes based on the known structure of the component proteins. The predictions were compared to the unpublished X-ray structures of the complexes. We describe here the motivations for launching CAPRI, the rules that we applied to select targets and run the experiment, and some conclusions that can already be drawn. The results stress the need for new scoring functions and for methods handling the conformation changes that were observed in some of the target systems. CAPRI has already been a powerful drive for the community of computational biologists who development docking algorithms. We hope that this issue of Proteins will also be of interest to the community of structural biologists, which we call upon to provide new targets for future rounds of CAPRI, and to all molecular biologists who view protein-protein recognition as an essential process.

The object of this work was to improve multiple sequence alignments using public-domain software and methods as far as possible. A method is described where the secondary structure of proteins is predicted and this information, coupled with a simplified description of the amino acids, is used to produce multiple sequence alignments. This method improved the accuracy of the resulting alignments by between 5 and 14% when compared with full sequence profile alignments (as scored against structural alignments). These improved alignments were used to predict the secondary structure of the sequences they contain. The resultant predictions were more accurate than those produced from less optimal alignments. An improvement of 6% for a three-state (helix, sheet and coil) prediction was observed when using the best alignment from the method presented here and the alignment obtained using sequence only. The method makes use of public domain software and all the associated files required to repeat the work are available from the primary author.

Fourteen models were constructed and analyzed for the comparative modeling section of Critical Assessment of Techniques for Protein Structure Prediction (CASP4). Sequence identity between each target and the best possible parent(s) ranged between 55 and 13%, and the root-mean-square deviation between model and target was from 0.8 to 17.9 A. In the fold recognition section, 10 of the 11 remote homologues were recognized. The modeling protocols are a combination of automated computer algorithms, 3D-JIGSAW (for comparative modeling) and 3D-PSSM (for fold recognition), with human intervention at certain critical stages. In particular, intervention is required to check superfamily assignment, best possible parents from which to model, sequence alignments to those parents and take-off regions for modeling variable regions. There now is a convergence of algorithms for comparative modeling and fold recognition, particularly in the region of remote homology.

As a form of Machine Learning the study of Inductive Logic Programming (ILP) is motivated by a central belief: relational description languages are better (in terms of accuracy and understandability) than propositional ones for certain real-world applications. This claim is investigated here for a particular application in structural molecular biology, that of constructing readable descriptions of the major protein folds. To the authors' knowledge Machine Learning has not previously been applied systematically to this task. In this application, the domain expert (third author) identified a natural divide between essentially propositional features and more structurally-oriented relational ones. The following null hypotheses are tested: 1) for a given ILP system (Progol) provision of relational background knowledge does not increase predictive accuracy, 2) a good propositional learning system (C5.0) without relational background knowledge will outperform Progol with relational background knowledge, 3) relational background knowledge does not produce improved explanatory insight. Null hypotheses 1) and 2) are both refuted on cross-validation results carried out over 20 of the most populated protein folds. Hypothesis 3 is refuted by demonstration of various insightful rules discovered only in the relationally-oriented learned rules.

Empirical residue–residue pair potentials are used to screen possible complexes for protein–protein dockings. A correct docking is defined as a complex with not more than 2.5 A root-mean-square distance from the known experimental structure. The complexes were generated by “ftdock” (Gabb et al. J Mol Biol 1997;272:106–120) that ranks using shape complementarity. The complexes studied were 5 enzyme-inhibitors and 2 antibody-antigens, starting from the unbound crystallographic coordinates, with a further 2 antibody-antigens where the antibody was from the bound crystallographic complex. The pair potential functions tested were derived both from observed intramolecular pairings in a database of nonhomologous protein domains, and from observed intermolecular pairings across the interfaces in sets of nonhomologous heterodimers and homodimers. Out of various alternate strategies, we found the optimal method used a mole-fraction calculated random model from the intramolecular pairings. For all the systems, a correct docking was placed within the top 12% of the pair potential score ranked complexes. A combined strategy was developed that incorporated “multidock,” a side-chain refinement algorithm (Jackson et al. J Mol Biol 1998;276:265–285). This placed a correct docking within the top 5 complexes for enzyme-inhibitor systems, and within the top 40 complexes for antibody–antigen systems. Proteins 1999;35:364–373. © 1999 Wiley-Liss, Inc.

Ten models were constructed for the comparative modeling section of the Critical Assessment of Techniques for Protein Structure Prediction-3 (CASP3). Sequence identity between each target and the best possible parent(s) ranged between 12% and 64%. The modeling protocol is a mixture of automated computer algorithms with human intervention at certain critical stages. In particular, intervention is required to check sequence alignments and the selection of parameters for various computer programs. Seven of the targets were constructed from single-parent templates, and three were constructed from multiple parents. The reasons for such a high ratio of modeling from single parents only are discussed. Models constructed from multiple parents were found to be more accurate than models constructed from single parents only. A novel loop-modeling algorithm is presented that consists of fragment database searches, several fragment libraries, and mean-field calculations on representative fragment candidates. Proteins Suppl 1999;3:47–54. © 1999 Wiley-Liss, Inc.

The docking of repressor proteins to DNA starting from the unbound protein and model-built DNA coordinates is modeled computationally. The approach was evaluated on eight repressor/DNA complexes that employed different modes for protein/ DNA recognition. The global search is based on a protein-protein docking algorithm that evaluates shape and electrostatic complementarity, which was modified to consider the importance of electrostatic features in DNA-protein recognition. Complexes were then ranked by an empirical score for the observed amino acid /nucleotide pairings (i.e., protein-DNA pair potentials) derived from a database of 20 protein/ DNA complexes. A good prediction had at least 65% of the correct contacts modeled. This approach was able to identify a good solution at rank four or better for three out of the eight complexes. Predicted complexes were filtered by a distance constraint based on experimental data defining the DNA footprint. This improved coverage to four out of eight complexes having a good model at rank four or better. The additional use of amino acid mutagenesis and phylogenetic data defining residues on the repressor resulted in between 2 and 27 models that would have to be examined to find a good solution for seven of the eight test systems. This study shows that starting with unbound coordinates one can predict three-dimensional models for protein/DNA complexes that do not involve gross conformational changes on association.

The BRCT domain (BRCA1 C-terminus), first identified in the breast cancer suppressor protein BRCA1, is an evolutionarily conserved protein-protein interaction region of approximately 95 amino acids found in a large number of proteins involved in DNA repair, recombination and cell cycle control. Here we describe the first three-dimensional structure and fold of a BRCT domain determined by X-ray crystallography at 3.2 A resolution. The structure has been obtained from the C-terminal region of the human DNA repair protein XRCC1, and comprises a four-stranded parallel beta-sheet surrounded by three alpha-helices, which form an autonomously folded domain. The compact XRCC1 structure explains the observed sequence homology between different BRCT motifs and provides a framework for modelling other BRCT domains. Furthermore, the established structure of an XRCC1 BRCT homodimer suggests potential protein-protein interaction sites for the complementary BRCT domain in DNA ligase III, since these two domains form a stable heterodimeric complex. Based on the XRCC1 BRCT structure, we have constructed a model for the C-terminal BRCT domain of BRCA1, which frequently is mutated in familial breast and ovarian cancer. The model allows insights into the effects of such mutations on the fold of the BRCT domain.

Information on the preparation and characterization of heavy-atom derivatives of protein crystals has been collected, either from the literature or directly from protein crystallographers, and assembled in the form of a heavy-atom data bank (HAD). The data bank contains coordinate data for the heavy-atom positions in a form that is compatible with the crystallographic data in the Brookhaven Protein Data Bank, together with a wealth of information on the crystallization conditions, the nature of the heavy-atom reagent and references to relevant publications. Some statistical information derived from the data bank, such as the most popular heavy-atom derivatives, is also included. The information can be directly accessed and should be useful to protein crystallographers seeking to improve their success in preparing heavy-atom derivatives for the methods of isomorphous replacement and anomalous dispersion. The World Wide Web address of HAD is http://www.icnet.uk/bmm/had.

Although it is well known that significant sequence similarity between proteins is reflected at the structural level, it is commonly assumed that any misaligned regions, as judged by the correct structure based alignment, are those where the local sequence identity is lower than the global. Recent studies have shown that this is not always the case and there can exist short stretches of high local identity which is not reflected in the structure based alignment. An analysis is presented of 290 pairs of homologous proteins with a view to quantifying the occurrence of these misleading local sequence alignments (MLSAs). It is found that such MLSAs are likely if the global sequence identity is less than 40% and can occur even when it is greater than 60%. The results have implications for automated homology modelling and also for the inference of function made by comparison.

Fold recognition methods aim to use the information in the known protein structures (the targets) to identify that the sequence of a protein of unknown structure (the probe) will adopt a known fold. This paper highlights that the structural similarities sought by these methods can be divided into two types: remote homologues and analogues. Homologues are the result of divergent evolution and often share a common function. We define remote homologues as those that are not easily detectable by sequence comparison methods alone. Analogues do not have a common ancestor and generally do not have a common function. Several sets of empirical matrices for residue substitution, secondary structure conservation and residue accessibility conservation have previously been derived from aligned pairs of remote homologues and analogues (Russell et al., J. Mol. Biol., 1997, 269, 423-439). Here a method for fold recognition, FOLDFIT, is introduced that uses these matrices to match the sequences, secondary structures and residue accessibilities of the probe and target. The approach is evaluated on distinct datasets of analogous and remotely homologous folds. The accuracy of FOLDFIT with the different matrices on the two datasets is contrasted to results from another fold recognition method (THREADER) and to searches using mutation matrices in the absence of any structural information. FOLDFIT identifies at top rank 12 out of 18 remotely homologous folds and five out of nine analogous folds. The average alignment accuracies for residue and secondary structure equivalencing are much higher for homologous folds (residue approximately 42%, secondary structure approximately 78%) than for analogues folds (approximately 12%, approximately 47%). Sequence searches alone can be successful for several homologues in the testing sets but nearly always fail for the analogues. These results suggest that the recognition of analogous and remotely homologous folds should be assessed separately. This study has implications for the development and comparative evaluation of fold recognition algorithms.

Two three-dimensional protein structural similarities accompanied by slight similarities in function are reported to highlight the present difficulties in discerning the relationship between structure and function. The similarity between the structures of uteroglobin and the CAP domain of haloalkane dehalogenase reveals a common four helix hydrophobic ligand binding motif. The similarity between the beta-barrel domains of glutaminyl tRNA synthetase and domain 1 of glutamine synthetase is accompanied by some similarity in the location of nucleotide binding sites and suggests a possible ancient domain common to these two proteins involved in the synthesis or binding of glutaminyl moieties. The problems raised by these two similarities in the structure-function 'twilight zone' are discussed.

The CASP blind trials (Critical Assessment of techniques for protein Structure Prediction) assess the accuracy of protein prediction that includes evaluation of comparative model building of protein structures. Comparative models of four proteins (T0001, T0003, T0017, and T0028) for CASP2 (held during 1996) were constructed using computer algorithms combined with visual inspection. Essentially the main-chain modelling involves construction of the target structure from rigid-body segments of homologues and loop fragments extracted from homologous and nonredundant databases. Side-chains were initially constructed by inheritance from the parent or from a rotamer library. Side-chain conformations were then refined using a novel mean field approach that includes solvation. Comparison of the models with the subsequently released X-ray structures identified the successes and limitations of our approach. The most problematic area is the quality of the sequence alignments between parent(s) and target. In this respect the overinterpretation of the conserved features within homologous families can be misleading. Several features of our approach have a positive effect on the accuracy of the models. For T0003, inspection correctly identified that a lower sequence identity parent provides the best framework for this model. Loop selection worked well where a homologous protein fragment was used, but that the use of nonredundant fragment library remains problematic for hinge movements and displacements in secondary structure elements relative to the parent. Side-chain refinement improved residue conformations relative to the initial model. Use of limited energy minimization improved the stereochemical quality of the model without increasing the RMS deviation. This study has identified methods that are effective and areas requiring further attention to improve model building by comparison. Proteins, Suppl. 1:59–67, 1997. © 1998 Wiley-Liss, Inc.

Crystallization of the 1:1 molecular complex between the beta-lactamase TEM-1 and the beta-lactamase inhibitory protein BLIP has provided an opportunity to put a stringent test on current protein-docking algorithms. Prior to the successful determination of the structure of the complex, nine laboratory groups were given the refined atomic coordinates of each of the native molecules. Other than the fact that BLIP is an effective inhibitor of a number of beta-lactamase enzymes (KI for TEM-1 approximately 100 pM) no other biochemical or structural data were available to assist the practitioners in their molecular docking. In addition, it was not known whether the molecules underwent conformational changes upon association or whether the inhibition was competitive or non-competitive. All six of the groups that accepted the challenge correctly predicted the general mode of association of BLIP and TEM-1.

Similar protein substructures can reveal a common function, suggest a distant evolutionary relationship or identify a folding motif adopted by proteins having no significant sequence or functional similarity (e.g. Orengo et al., 1994). In this correspondence, we examine the similarities between the three-dimensional (3-D) structures of catalase and the calycin (Flower, 1993) structure superfamily and suggest a putative binding site in catalase. We have performed an all-against-all structural comparison of a database of author-assigned protein domains (Islam et al., 1995) using the STAMP algorithm (Russell and Barton, 1992). During this study, we found a significant similarity between the 3-D structures of calycins (including lipocalins, fatty acidbinding proteins and avidins; Flower, 1993) and the catalase domain (residues 72-318 in beef catalase). These structures have a similar eight-stranded P-barrel with sequential strands forming P-hairpins. We considered two diverse representative members of the calycin family: avidin (Livnah et al., 1993) and retinol-binding protein (RBP) (Newcomer, 1984), a lipocalin. Both proteins show a significant degree of similarity to catalase: 55 and 62 Ca atoms from RBP and avidin respectively can be superimposed on to catalase with r.m.s. deviations of 2.1 A. Fifty-two Ca atoms from RBP can be superimposed onto avidin with a similar r.m.s. deviation. A structure-based sequence alignment of these three structures is shown in Figure 1. 3-D structures of RBP, avidin and catalase are shown in Figure 2 in the same orientation, with equivalent residues highlighted. p*-Barrels can be classified by two numbers, a and b (a, b) where a is the number of fi-strands they contain and b is an offset value (McLachlan, 1979). Under this classification, catalase is more similar to avidin in architecture: both have (8,10) P-barrels, whereas the RBP barrel is (8,12) (Murzin et al., 1995). The similarity between catalase and RBP has been noted briefly (Newcomer, 1984; Holm and Sander, 1993).

Problems concerned with learning the relationships between molecular structure and activity have been important test-beds for Inductive Logic programming (ILP) systems. In this paper we examine these applications and empirically evaluate the extent to which a first-order representation was required. We compared ILP theories with those constructed using standard linear regression and a decision-tree learner on a series of progressively more difficult problems. When a propositional encoding is feasible for the feature-based algorithms, we show that such algorithms are capable of matching the predictive accuracies of an ILP theory. However, as the complexity of the compounds considered increased, propositional encodings becomes intractable. In such cases, our results show that ILP programs can still continue to construct accurate, understandable theories. Based on this evidence, we propose future work to realise fully the potential of ILP in structure-activity problem.

A common step in pharmaceutical development is the formation of a quantitative structure-activity relationship *(QSAR) to model an exploratory series of compounds. A QSAR generalizes how the structure (shape) of a compound relates to its biological activity. A comparative study was carried out of six artificial intelligence and traditional algorithms for modeling QSAR's: GOLEM, CART, and MS from symbolic machine learning; back-propagation from neural networks; and linear regression and nearest-neighbor from traditional statistics. Two test case problems were studied: the inhibition of Escherichia coli dihydrofolate reductase (DHFR) by pyrimidines, and the inhibition of ratlmouse tumor DHFR by triazines. It was found that there was no significant statistical difference between the methods in terms of their ability to rank unseen compounds by activity. However, symbolic machine learning methods, in particular relational ones, were found to generate rules that provided insight into the stereochemistry of com...

3,4-Dihydro-2(1H)-isoquinolinecarboxamidinium sulfate, 2C 10 H 14 N + 3. SO 4 2- , M r =448.5, monoclinic, C2/c, a=27.214 (10), b=6.552 (1), c=13.241 (1) A, β=113.35 (2) o , V=2168 (1) A 3 , Z=4, D x =1.375 g cm -3 , λ(Mo Kα)=0.7107 A, μ=1.81 cm -3 , F(000)=952, room temperature, final R=0.042 for 2446 unique reflections [1661 observed with I>3σ(I)]. The relative orientations of the two h atoms on C(4) of debrisoquin, one of which is enzymatically replaced by a hydroxyl group, is discussed

Bacteria produce an array of glycan-based structures including capsules, lipo-oligosaccharide and glycosylated proteins, which are invariably cell-surface-located. For pathogenic bacteria, such structures are involved in diverse roles in the life cycle of the bacterium, including adhesion, colonization, avoidance of predation and interactions with the immune system. Compared with eukaryotes, bacteria produce huge combinatorial variations of glycan structures, which, coupled to the lack of genetic data, has previously hampered studies on bacterial glycans and their role in survival and pathogenesis. The advent of genomics in tandem with rapid technological improvements in MS analysis has opened a new era in bacterial glycomics. This has resulted in a rich source of novel glycan structures and new possibilities for glycoprospecting and glycoengineering. However, assigning genetic information in predicted glycan biosynthetic pathways to the overall structural information is complex. Bioinformatic analysis is required, linked to systematic mutagenesis and functional analysis of individual genes, often from diverse biosynthetic pathways. This must then be related back to structural analysis from MS or NMR spectroscopy. To aid in this process, systems level analysis of the multiple datasets can be used to make predictions of gene function that can then be confirmed experimentally. The present paper exemplifies these advances with reference to the major gastrointestinal pathogen Campylobacter jejuni.

The 3D-Dock suite of programs has been used to make predictions for the seven targets in rounds 1 and 2 of the CAPRI method evaluation exercise. Some correct contacts were obtained in at least one prediction for four of seven targets. Target 06 was predicted very well, with an RMSD of the ligand after superimposition of the receptor of only 0.77 A. We investigate the performance of the various stages of the method, with the aim of finding where improvements need to be made, and in particular whether the manual interventions that were made were essential, and whether results of the level of accuracy obtained for target 06 may be expected with confidence.

A technique of machine learning, inductive logic programming implemented in the program GOLEM, has been applied to three problems in structural molecular biology. These problems are: the prediction of protein secondary structure; the identification of rules governing the arrangement of β-sheets strands in the tertiary folding of proteins; and the modelling of a quantitative structure activity relationship (QSAR) of a series of drugs. For secondary structure prediction and the QSAR, GOLEM yielded predictions comparable with contemporary approaches including neural networks. Rules for β-strand arrangement are derived and it is planned to contrast their accuracy with those obtained by human inspection. In all three studies GOLEM discovered rules that provided insight into the stereochemistry of the system. We conclude machine leaning used together with human intervention will provide a powerful tool to discover patterns in biological sequences and structures.

A three-dimensional molecular template has been generated for substrates of human debrisoquine 4-hydroxylase cytochrome P450 (CYP2D6). This template defines the stereochemical requirements for CYP2D6 substrates in terms of the volume occupied and positions of key atoms. The modelling was based on the X-ray crystallographic coordinates of the location of the attacked C5 atom of camphor in relation to the haem in cytochrome P450 cam. Interactive molecular graphics combined with energy calculations were used to identify allowed conformers to superpose known CYP2D6 substrates to yield a molecular template. This model takes into account the site of attack of the known substrates and the requirement for a protonated nitrogen atom to interact with an anion site of the protein. A nitrogen-anion distance of between 2.5 and 4.5 A was allowed for the interaction. The substrates modelled were cardiovascular drugs (debrisoquine, sparteine, guanoxan and perhexiline), beta-adrenergic blocking agents (bufuralol and propranolol), tricyclic anti-depressants (desipramine, amitriptyline and nortriptyline) and other miscellaneous compounds (phenformin, methoxy-amphetamine, codeine and dextromethorphan). The template generated in this manner was then used to determine the likelihood that certain other compounds were substrates for CYP2D6. A carcinogenic protein pyrolysate product, 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP), did not fit the template and is therefore unlikely to be activated by this enzyme. A potent carcinogen in tobacco smoke, 4-(N-methyl-N-nitrosamino)-1-(3-pyridyl)-1-butanone (NNK), fitted the template but could not be modelled to form a favourable nitrogen-anion interaction. Experimental substrate competition studies also showed that NNK is unlikely to be a CYP2D6 substrate. It was also shown that the widely used drug for treatment of breast cancer, trans-1-(4-beta-dimethylaminoethoxyphenyl)-,2-diphenyl-1-ene (tamoxifen), did not fit the molecular template and is unlikely to be metabolized by CYP2D6. Coordinates of the template are available.

Background Microarray experimentation requires the application of complex analysis methods as well as the use of non-trivial computer technologies to manage the resultant large data sets. This, together with the proliferation of tools and techniques for microarray data analysis, makes it very challenging for a laboratory scientist to keep up-to-date with the latest developments in this field. Our aim was to develop a distributed e-support system for microarray data analysis and management. Results EMAAS (Extensible MicroArray Analysis System) is a multi-user rich internet application (RIA) providing simple, robust access to up-to-date resources for microarray data storage and analysis, combined with integrated tools to optimise real time user support and training. The system leverages the power of distributed computing to perform microarray analyses, and provides seamless access to resources located at various remote facilities. The EMAAS framework allows users to import microarray data from several sources to an underlying database, to pre-process, quality assess and analyse the data, to perform functional analyses, and to track data analysis steps, all through a single easy to use web portal. This interface offers distance support to users both in the form of video tutorials and via live screen feeds using the web conferencing tool EVO. A number of analysis packages, including R-Bioconductor and Affymetrix Power Tools have been integrated on the server side and are available programmatically through the Postgres-PLR library or on grid compute clusters. Integrated distributed resources include the functional annotation tool DAVID, GeneCards and the microarray data repositories GEO, CELSIUS and MiMiR. EMAAS currently supports analysis of Affymetrix 3' and Exon expression arrays, and the system is extensible to cater for other microarray and transcriptomic platforms. Conclusion EMAAS enables users to track and perform microarray data management and analysis tasks through a single easy-to-use web application. The system architecture is flexible and scalable to allow new array types, analysis algorithms and tools to be added with relative ease and to cope with large increases in data volume.

1. DSC has high accuracy. It has a prediction accuracy of 70.1% (per residue) on a standard set of 126 proteins. This percentage was confirmed by the recent CASP2 blind prediction challenge (see below). 2. DSC is based on simple linear statistics. Existing highaccuracy prediction methods are 'black-box' predictors based on complex non-linear statistics [e.g. neural networks in PHD (Rost and Sander, 1993) and nearestneighbour methods in NNSSP (Salamov and Solovyev, 1995)]. 3. DSC's accuracy is predictable a priori. This permits evaluation of the utility of the prediction, e.g. it is possible accurately to predict chains that have a mean accuracy of >80%. 4. DSC source code is freely available (in C). This allows DSC to be easily ported to other systems and to be used on proprietary sequences (e.g. DSC is running in-house at Glaxo Welcome pic). 5. DSC has a prediction server (see below). This server can generate its own sequence alignments.

A machine learning program, GOLEM, has been applied to two problems: (1) the prediction of protein secondary structure from sequence and (2) modelling a quantitative structure-activity relationship in drug design. GOLEM takes as input observations and combines them with background knowledge of chemistry to yield rules expressed as stereochemical principles for prediction. The secondary structure prediction was explored on the alpha/alpha class of proteins; on an unrelated test set it yielded 81% accuracy. The rules from GOLEM defined patterns of residues forming alpha-helices. The system studied for drug design was the activities of trimethoprim analogues binding to E. coli dihydrofolate reductase. The GOLEM rules were a better model than standard regression approaches. More importantly, these rules described the chemical properties of the enzyme-binding site that were in broad agreement with the crystallographic structure.