SUMMARY: Many eukaryotic proteins are modified by N-glycans. Liquid chromatography (ultra-performance -UPLC and high-performance-HPLC) coupled with mass spectrometry (MS) is conventionally used to characterize N-glycan structures. Software can automatically assign glycan structures by matching their observed retention times and masses with standardized values in reference databases. However, more precise confirmation of N-glycan structures can be derived using exoglycosidases, enzymes that remove specific monosaccharides from glycans. Exoglycosidase removal of monosaccharides results in signature peak shifts, in both UPLC and MS1, yielding an effective way to verify N-glycan structure with high detail (down to the position and isomeric linkage of each monosaccharide). Because manual interpretation of exoglycosidase data is complex and time consuming, we developed GlycanAnalyzer, a web application that pattern matches N-glycan peak shifts following exoglycosidase digestion and automates structure assignments. GlycanAnalyzer significantly improves assignment accuracy over other auto-assignment methods on tests with a monoclonal antibody and four glycan standards (100% versus 82% for the next best software). By automating data interpretation, GlycanAnalyzer enables the easier use of exoglycosidases to precisely define N-glycan structure. AVAILABILITY AND IMPLEMENTATION: http://glycananalyzer.neb.com. Datasets available online. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Bioinformatics (2019) doi: 10.1093/bioinformatics/bty681, 35, 688–690. In the original article, author Tasha José’s first and surname were originally transposed. This has now been corrected.

Summary: GlycoStore is a curated chromatographic, electrophoretic and mass-spectrometry composition database of N-, O-, glycosphingolipid (GSL) glycans and free oligosaccharides associated with a range of glycoproteins, glycolipids and biotherapeutics. The database is built on publicly available experimental datasets from GlycoBase developed in the Oxford Glycobiology Institute and then the National Institute for Bioprocessing Research and Training (NIBRT). It has now been extended to include recently published and in-house data collections from the Bioprocessing Technology Institute (BTI) A*STAR, Macquarie University and Ludger Ltd. GlycoStore provides access to approximately 850 unique glycan structure entries supported by over 8500 retention positions determined by: (i) hydrophilic interaction chromatography (HILIC) ultra-high performance liquid chromatography (U/HPLC) and reversed phase (RP)-U/HPLC with fluorescent detection; (ii) porous graphitized carbon (PGC) chromatography in combination with ESI-MS/MS detection; and (iii) capillary electrophoresis with laser induced fluorescence detection (CE-LIF). GlycoStore enhances many features previously available in GlycoBase while addressing the limitations of the data collections and model of this popular resource. GlycoStore aims to support detailed glycan analysis by providing a resource that underpins current workflows. It will be regularly updated by expert annotation of published data and data obtained from the project partners. Availability and implementation: http://www.glycostore.org. Supplementary information: Supplementary data are available at Bioinformatics online.

MOTIVATION: The behavior of a protein is encoded in its sequence, which can be used to predict distinct features such as secondary structure, intrinsic disorder or amphipathicity. Integrating these and other features can help explain the context-dependent behavior of proteins. However, most tools focus on a single aspect, hampering a holistic understanding of protein structure. Here, we present Fast Estimator of Latent Local Structure (FELLS) to visualize structural features from the protein sequence. FELLS provides disorder, aggregation and low complexity predictions as well as estimated local propensities including amphipathicity. A novel fast estimator of secondary structure (FESS) is also trained to provide a fast response. The calculations required for FELLS are extremely fast and suited for large-scale analysis while providing a detailed analysis of difficult cases. AVAILABILITY AND IMPLEMENTATION: The FELLS web server is available from URL: http://protein.bio.unipd.it/fells/ . The server also exposes RESTful functionality allowing programmatic prediction requests. An executable version of FESS for Linux can be downloaded from URL: protein.bio.unipd.it/download/. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Machine learning methods are becoming increasingly popular to predict protein features from sequences. Machine learning in bioinformatics can be powerful but carries also the risk of introducing unexpected biases, which may lead to an overestimation of the performance. This article espouses a set of guidelines to allow both peer reviewers and authors to avoid common machine learning pitfalls. Understanding biology is necessary to produce useful data sets, which have to be large and diverse. Separating the training and test process is imperative to avoid over-selling method performance, which is also dependent on several hidden parameters. A novel predictor has always to be compared with several existing methods, including simple baseline strategies. Using the presented guidelines will help nonspecialists to appreciate the critical issues in machine learning.

MOTIVATION: Intrinsically disordered regions are key for the function of numerous proteins. Due to the difficulties in experimental disorder characterization, many computational predictors have been developed with various disorder flavors. Their performance is generally measured on small sets mainly from experimentally solved structures, e.g. Protein Data Bank (PDB) chains. MobiDB has only recently started to collect disorder annotations from multiple experimental structures. RESULTS: MobiDB annotates disorder for UniProt sequences, allowing us to conduct the first large-scale assessment of fast disorder predictors on 25 833 different sequences with X-ray crystallographic structures. In addition to a comprehensive ranking of predictors, this analysis produced the following interesting observations. (i) The predictors cluster according to their disorder definition, with a consensus giving more confidence. (ii) Previous assessments appear over-reliant on data annotated at the PDB chain level and performance is lower on entire UniProt sequences. (iii) Long disordered regions are harder to predict. (iv) Depending on the structural and functional types of the proteins, differences in prediction performance of up to 10% are observed. AVAILABILITY: The datasets are available from Web site at URL: http://mobidb.bio.unipd.it/lsd. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Protein inter-residue contact maps provide a translation and rotation invariant topological representation of a protein. They can be used as an intermediary step in protein structure predictions. However, the prediction of contact maps represents an unbalanced problem as far fewer examples of contacts than non-contacts exist in a protein structure.In this study we explore the possibility of completely eliminating the unbalanced nature of the contact map prediction problem by predicting real-value distances between residues. Predicting full inter-residue distance maps and applying them in protein structure predictions has been relatively unexplored in the past. RESULTS: We initially demonstrate that the use of native-like distance maps is able to reproduce 3D structures almost identical to the targets, giving an average RMSD of 0.5Å. In addition, the corrupted physical maps with an introduced random error of ±6Å are able to reconstruct the targets within an average RMSD of 2Å.After demonstrating the reconstruction potential of distance maps, we develop two classes of predictors using two-dimensional recursive neural networks: an ab initio predictor that relies only on the protein sequence and evolutionary information, and a template-based predictor in which additional structural homology information is provided. We find that the ab initio predictor is able to reproduce distances with an RMSD of 6Å, regardless of the evolutionary content provided. Furthermore, we show that the template-based predictor exploits both sequence and structure information even in cases of dubious homology and outperforms the best template hit with a clear margin of up to 3.7Å.Lastly, we demonstrate the ability of the two predictors to reconstruct the CASP9 targets shorter than 200 residues producing the results similar to the state of the machine learning art approach implemented in the Distill server. CONCLUSIONS: The methodology presented here, if complemented by more complex reconstruction protocols, can represent a possible path to improve machine learning algorithms for 3D protein structure prediction. Moreover, it can be used as an intermediary step in protein structure predictions either on its own or complemented by NMR restraints.

BACKGROUND: Intrinsic protein disorder is becoming an increasingly important topic in protein science. During the last few years, intrinsically disordered proteins (IDPs) have been shown to play a role in many important biological processes, e.g. protein signalling and regulation. This has sparked a need to better understand and characterize different types of IDPs, their functions and roles. Our recently published database, MobiDB, provides a centralized resource for accessing and analysing intrinsic protein disorder annotations. RESULTS: Here, we present a thorough description and analysis of the data made available by MobiDB, providing descriptive statistics on the various available annotation sources. Version 1.2.1 of the database contains annotations for ca. 4,500,000 UniProt sequences, covering all eukaryotic proteomes. In addition, we describe a novel consensus annotation calculation and its related weighting scheme. The comparison between disorder information sources highlights how the MobiDB consensus captures the main features of intrinsic disorder and correlates well with manually curated datasets. Finally, we demonstrate the annotation of 13 eukaryotic model organisms through MobiDB's datasets, and of an example protein through the interactive user interface. CONCLUSIONS: MobiDB is a central resource for intrinsic disorder research, containing both experimental data and predictions. In the future it will be expanded to include additional information for all known proteins.

MOTIVATION: Disordered protein regions are key to the function of numerous processes within an organism and to the determination of a protein's biological role. The most common source for protein disorder annotations, DisProt, covers only a fraction of the available sequences. Alternatively, the Protein Data Bank (PDB) has been mined for missing residues in X-ray crystallographic structures. Herein, we provide a centralized source for data on different flavours of disorder in protein structures, MobiDB, building on and expanding the content provided by already existing sources. In addition to the DisProt and PDB X-ray structures, we have added experimental information from NMR structures and five different flavours of two disorder predictors (ESpritz and IUpred). These are combined into a weighted consensus disorder used to classify disordered regions into flexible and constrained disorder. Users are encouraged to submit manual annotations through a submission form. MobiDB features experimental annotations for 17 285 proteins, covering the entire PDB and predictions for the SwissProt database, with 565 200 annotated sequences. Depending on the disorder flavour, 6-20% of the residues are predicted as disordered. AVAILABILITY: The database is freely available at http://mobidb.bio.unipd.it/. CONTACT: [email protected].

MOTIVATION: Electrostatic calculations are an important tool for deciphering many functional mechanisms in proteins. Generalized Born (GB) models offer a fast and convenient computational approximation over other implicit solvent-based electrostatic models. Here we present a novel GB-based web server, using the program Bluues, to calculate numerous electrostatic features including pKa-values and surface potentials. The output is organized allowing both experts and beginners to rapidly sift the data. A novel feature of the Bluues server is that it explicitly allows to find electrostatic differences between wild-type and mutant structures. AVAILABILITY: The Bluues server, examples and extensive help files are available for non-commercial use at URL: http://protein.bio.unipd.it/bluues/.

MOTIVATION: Intrinsically disordered regions are key for the function of numerous proteins, and the scant available experimental annotations suggest the existence of different disorder flavors. While efficient predictions are required to annotate entire genomes, most existing methods require sequence profiles for disorder prediction, making them cumbersome for high-throughput applications. RESULTS: In this work, we present an ensemble of protein disorder predictors called ESpritz. These are based on bidirectional recursive neural networks and trained on three different flavors of disorder, including a novel NMR flexibility predictor. ESpritz can produce fast and accurate sequence-only predictions, annotating entire genomes in the order of hours on a single processor core. Alternatively, a slower but slightly more accurate ESpritz variant using sequence profiles can be used for applications requiring maximum performance. Two levels of prediction confidence allow either to maximize reasonable disorder detection or to limit expected false positives to 5%. ESpritz performs consistently well on the recent CASP9 data, reaching a S(w) measure of 54.82 and area under the receiver operator curve of 0.856. The fast predictor is four orders of magnitude faster and remains better than most publicly available CASP9 methods, making it ideal for genomic scale predictions. CONCLUSIONS: ESpritz predicts three flavors of disorder at two distinct false positive rates, either with a fast or slower and slightly more accurate approach. Given its state-of-the-art performance, it can be especially useful for high-throughput applications. AVAILABILITY: Both a web server for high-throughput analysis and a Linux executable version of ESpritz are available from: http://protein.bio.unipd.it/espritz/.

MOTIVATION: Repeat proteins form a distinct class of structures where folding is greatly simplified. Several classes have been defined, with solenoid repeats of periodicity between ca. 5 and 40 being the most challenging to detect. Such proteins evolve quickly and their periodicity may be rapidly hidden at sequence level. From a structural point of view, finding solenoids may be complicated by the presence of insertions or multiple domains. To the best of our knowledge, no automated methods are available to characterize solenoid repeats from structure. RESULTS: Here we introduce RAPHAEL, a novel method for the detection of solenoids in protein structures. It reliably solves three problems of increasing difficulty: (1) recognition of solenoid domains, (2) determination of their periodicity and (3) assignment of insertions. RAPHAEL uses a geometric approach mimicking manual classification, producing several numeric parameters that are optimized for maximum performance. The resulting method is very accurate, with 89.5% of solenoid proteins and 97.2% of non-solenoid proteins correctly classified. RAPHAEL periodicities have a Spearman correlation coefficient of 0.877 against the manually established ones. A baseline algorithm for insertion detection in identified solenoids has a Q(2) value of 79.8%, suggesting room for further improvement. RAPHAEL finds 1931 highly confident repeat structures not previously annotated as solenoids in the Protein Data Bank records.

MOTIVATION: Residue interaction networks (RINs) have been used in the literature to describe the protein 3D structure as a graph where nodes represent residues and edges physico-chemical interactions, e.g. hydrogen bonds or van-der-Waals contacts. Topological network parameters can be calculated over RINs and have been correlated with various aspects of protein structure and function. Here we present a novel web server, RING, to construct physico-chemically valid RINs interactively from PDB files for subsequent visualization in the Cytoscape platform. The additional structure-based parameters secondary structure, solvent accessibility and experimental uncertainty can be combined with information regarding residue conservation, mutual information and residue-based energy scoring functions. Different visualization styles are provided to facilitate visualization and standard plugins can be used to calculate topological parameters in Cytoscape. A sample use case analyzing the active site of glutathione peroxidase is presented. AVAILABILITY: The RING server, supplementary methods, examples and tutorials are available for non-commercial use at URL: http://protein.bio.unipd.it/ring/.

MOTIVATION: MOBI is a web server for the identification of structurally mobile regions in NMR protein ensembles. It provides a binary mobility definition that is analogous to the commonly used definition of intrinsic disorder in X-ray crystallographic structures. At least three different use cases can be envisaged: (i) visualization of NMR mobility for structural analysis; (ii) definition of regions for reliable comparative modelling in protein structure prediction and (iii) definition of mobility in analogy to intrinsic disorder. MOBI uses structural superposition and local conformational differences to derive a robust binary mobility definition that is in excellent agreement with the manually curated definition used in the CASP8 experiment for intrinsic disorder in NMR structure. The output includes mobility-coloured PDB files, mobility plots and a FASTA formatted sequence file summarizing the mobility results. AVAILABILITY: The MOBI server and supplementary methods are available for non-commercial use at URL: http://protein.bio.unipd.it/mobi/.

BACKGROUND: Proteins, especially larger ones, are often composed of individual evolutionary units, domains, which have their own function and structural fold. Predicting domains is an important intermediate step in protein analyses, including the prediction of protein structures. RESULTS: We describe novel systems for the prediction of protein domain boundaries powered by Recursive Neural Networks. The systems rely on a combination of primary sequence and evolutionary information, predictions of structural features such as secondary structure, solvent accessibility and residue contact maps, and structural templates, both annotated for domains (from the SCOP dataset) and unannotated (from the PDB). We gauge the contribution of contact maps, and PDB and SCOP templates independently and for different ranges of template quality. We find that accurately predicted contact maps are informative for the prediction of domain boundaries, while the same is not true for contact maps predicted ab initio. We also find that gap information from PDB templates is informative, but, not surprisingly, less than SCOP annotations. We test both systems trained on templates of all qualities, and systems trained only on templates of marginal similarity to the query (less than 25% sequence identity). While the first batch of systems produces near perfect predictions in the presence of fair to good templates, the second batch outperforms or match ab initio predictors down to essentially any level of template quality.We test all systems in 5-fold cross-validation on a large non-redundant set of multi-domain and single domain proteins. The final predictors are state-of-the-art, with a template-less prediction boundary recall of 50.8% (precision 38.7%) within +/- 20 residues and a single domain recall of 80.3% (precision 78.1%). The SCOP-based predictors achieve a boundary recall of 74% (precision 77.1%) again within +/- 20 residues, and classify single domain proteins as such in over 85% of cases, when we allow a mix of bad and good quality templates. If we only allow marginal templates (max 25% sequence identity to the query) the scores remain high, with boundary recall and precision of 59% and 66.3%, and 80% of all single domain proteins predicted correctly. CONCLUSION: The systems presented here may prove useful in large-scale annotation of protein domains in proteins of unknown structure. The methods are available as public web servers at the address: http://distill.ucd.ie/shandy/ and we plan on running them on a multi-genomic scale and make the results public in the near future.

BACKGROUND: We describe Distill, a suite of servers for the prediction of protein structural features: secondary structure; relative solvent accessibility; contact density; backbone structural motifs; residue contact maps at 6, 8 and 12 Angstrom; coarse protein topology. The servers are based on large-scale ensembles of recursive neural networks and trained on large, up-to-date, non-redundant subsets of the Protein Data Bank. Together with structural feature predictions, Distill includes a server for prediction of Calpha traces for short proteins (up to 200 amino acids). RESULTS: The servers are state-of-the-art, with secondary structure predicted correctly for nearly 80% of residues (currently the top performance on EVA), 2-class solvent accessibility nearly 80% correct, and contact maps exceeding 50% precision on the top non-diagonal contacts. A preliminary implementation of the predictor of protein Calpha traces featured among the top 20 Novel Fold predictors at the last CASP6 experiment as group Distill (ID 0348). The majority of the servers, including the Calpha trace predictor, now take into account homology information from the PDB, when available, resulting in greatly improved reliability. CONCLUSION: All predictions are freely available through a simple joint web interface and the results are returned by email. In a single submission the user can send protein sequences for a total of up to 32k residues to all or a selection of the servers. Distill is accessible at the address: http://distill.ucd.ie/distill/.

BACKGROUND: Protein topology representations such as residue contact maps are an important intermediate step towards ab initio prediction of protein structure. Although improvements have occurred over the last years, the problem of accurately predicting residue contact maps from primary sequences is still largely unsolved. Among the reasons for this are the unbalanced nature of the problem (with far fewer examples of contacts than non-contacts), the formidable challenge of capturing long-range interactions in the maps, the intrinsic difficulty of mapping one-dimensional input sequences into two-dimensional output maps. In order to alleviate these problems and achieve improved contact map predictions, in this paper we split the task into two stages: the prediction of a map's principal eigenvector (PE) from the primary sequence; the reconstruction of the contact map from the PE and primary sequence. Predicting the PE from the primary sequence consists in mapping a vector into a vector. This task is less complex than mapping vectors directly into two-dimensional matrices since the size of the problem is drastically reduced and so is the scale length of interactions that need to be learned. RESULTS: We develop architectures composed of ensembles of two-layered bidirectional recurrent neural networks to classify the components of the PE in 2, 3 and 4 classes from protein primary sequence, predicted secondary structure, and hydrophobicity interaction scales. Our predictor, tested on a non redundant set of 2171 proteins, achieves classification performances of up to 72.6%, 16% above a base-line statistical predictor. We design a system for the prediction of contact maps from the predicted PE. Our results show that predicting maps through the PE yields sizeable gains especially for long-range contacts which are particularly critical for accurate protein 3D reconstruction. The final predictor's accuracy on a non-redundant set of 327 targets is 35.4% and 19.8% for minimum contact separations of 12 and 24, respectively, when the top length/5 contacts are selected. On the 11 CASP6 Novel Fold targets we achieve similar accuracies (36.5% and 19.7%). This favourably compares with the best automated predictors at CASP6. CONCLUSION: Our final system for contact map prediction achieves state-of-the-art performances, and may provide valuable constraints for improved ab initio prediction of protein structures. A suite of predictors of structural features, including the PE, and PE-based contact maps, is available at http://distill.ucd.ie.