Scalable web services for the PSIPRED Protein Analysis Workbench.
Buchan DW, Minneci F, Nugent TC, Bryson K, Jones DT. · 2013
Read Scalable web services for the PSIPRED Protein Analysis Workbench. by Buchan DW, Minneci F, Nugent TC, Bryson K, Jones DT. (2013) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3692098.txt.
Introduction
Buchan and colleagues describe the PSIPRED Protein Analysis Workbench, a single web portal that gathers the UCL group's protein prediction tools in one place. Instead of visiting separate services, users run the group's methods through one streamlined interface and explore results in a richer display.
The update also tackles scale. Predictions can be submitted through SOAP and XML-RPC web services, and new Hadoop packages let the heavier jobs run across clusters. Everything is offered free through the group's website. It is aimed at researchers who need protein structure and function predictions for many sequences at once, without installing or maintaining the software themselves.
Catalog record
| CREATOR | Buchan DW, Minneci F, Nugent TC, Bryson K, Jones DT. |
|---|---|
| YEAR | 2013 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Scalable web services for the PSIPRED Protein Analysis Workbench By Daniel W A Buchan, Federico Minneci, Tim C O Nugent, Kevin Bryson, David T Jones Published in: Nucleic Acids Research DOI: 10.1093/nar/gkt381 ABSTRACT Abstract Here, we present the new UCL Bioinformatics Group’s PSIPRED Protein Analysis Workbench. The Workbench unites all of our previously available analysis methods into a single web-based framework. The new web portal provides a greatly streamlined user interface with a number of new features to allow users to better explore their results. We offer a number of additional services to enable computationally scalable execution of our prediction methods; these include SOAP and XML-RPC web server access and new HADOOP packages. All software and services are available via the UCL Bioinformatics Group website at http://bioinf.cs.ucl.ac.uk/. INTRODUCTION Contemporary computer science is increasingly being defined as the era of ‘Big Data’, as near-unimaginable torrents of data are produced by a great number of human endeavours. This is no less true of Biology. Projects as diverse as ENCODE (1), the 1000 genomes project (2) and SOL-100 (3) amply demonstrate the current and accelerating creation of biological data. All of which is in need of timely, efficient analysis and annotation. The PSIPRED web server aggregates a number of protein annotation tools and provides services or software to allow users to perform truly scalable biological analyses. We have recently rationalized the available methods into two primary homepages—The Protein Analysis Workbench and The Protein Structure Workbench, dealing with protein sequence and protein structure annotation, respectively. We have recently updated a number of the underlying analysis methods and also now provide programmatic access to the web server via SOAP or XML-RPC. As always, we offer the majority of the analysis methods as software downloads, and in addition to this, we now provide HADOOP (4) packages for our most commonly used methods. ALGORITHMS AND SOFTWARE The PSIPRED workbench makes available the following sequence and structure annotation methods: PSIPRED, GenTHREADER, pGenTHREADER, pDomTHREADER, MEMSAT-SVM/MEMSAT3, MEMPACK, BioSerf, MetSite, HSPred, DISOPRED2, DomPred and FFPred. Here, we give a brief overview of the new or updated algorithms available via the PSIPRED workbench; for those algorithms that have not seen major updates, we refer the reader to our previous 2010 Web Server article (5) or seminal publications for particular methods (6). BioSerf v2.0 BioSerf is a fully automated homology modelling protocol, which performed well in the 2008, 2010 and 2012 CASP (7,8) experiments. The BioSerf method has been substantially changed in the last year and no longer makes use of the FRAGFOLD (9) de novo modelling software. BioSerf 2 now begins by simultaneously running PSIBLAST (10) against the FASTA PDB (11), running HHSearch (12) against the PDB and also running pGenTHREADER (13). For each method, a list of the best scoring PDB chains is compiled, and a homology model is built for each. pGenTHREADER and HHSearch natively output models, whereas MODELLER (14) is used to build models against the PSIBLAST results. All of the models are then jointly pooled and assessed by calculating the TM-SCORE (15) in an all-against-all manner. This matrix is used to cluster the models into similar sets, and the 10 largest and tightest scoring clusters are then selected, and a PDB model is selected as the cluster representative. The alignments for the top 10 representative models are retrieved from the initial PSIBLAST, HHSearch and pGenTHREADER runs and are then used as input for a final MODELLER job to produce the prediction model. Users of the PSIPRED web server will need a valid MODELLER key to access this prediction method. Figure 1 presents an overview of the process. [Figure: Flowchart of the BioSerf2 automated homology modelling protocol. Incoming query sequences are independently matched to PDB chains using PSIBLAST, pGenTHREADER and HH. The three sets of models produced are then compared by the TMJury process, which produces up to 10 candidate homologous structures. These structures and their alignments to the query sequence are used as input for MODELLER to produce a single final structure.] FFPred v2.0 FFPred (16) performs feature-based function prediction and assigns Gene Ontology (GO) classes to query sequences using support vector machines (SVMs), representing each GO class. The method has been extensively updated and re-trained making use of updated versions of all external feature prediction software (manuscript accepted). Analysis proceeds by annotating a query sequence with a wide range of features, such as secondary structure elements or disordered regions; this information is exploited by each SVM. SVM outputs are converted into posterior probabilities associated to the corresponding annotations using a standard method (17), and they are considered positive predictions whenever the probability is >0.5. The size of the FFPred vocabulary has now increased to include 442 GO classes, 96 from the molecular function domain and 346 in the biological process domain. Also, each GO prediction is assigned a reliability label of either ‘high’ or ‘low’ according to the classifier’s performance in terms of Matthew’s correlation coefficient, sensitivity, specificity and prediction; the 200 best-performing SVMs are assigned ‘high’ reliability, whereas the remainder are assigned ‘low’. The main output of FFPred consists of a tabulated list of the predicted GO classes providing the GO class code, GO description, GO ontology domain, the posterior probability of the prediction and the reliability level of the SVM. All predictions of GO classes with a ‘high’ reliability level are listed first, whereas predictions of GO classes with ‘low’ reliability are listed against a red background. A summary of the features predicted for the query sequence, which were used for the SVM analysis, is also made available. Finally, users can download two files, one containing the final list of positive predictions, and another one including the posterior probabilities for all of the 442 GO classes. MEMSAT-SVM The support vector machine-based transmembrane topology predictor MEMSAT-SVM (18) has been updated and is now capable of automatically identifying pore-lining regions in transmembrane proteins from sequence information alone. An SVM classifier was trained using sequence profile data by labelling pore-lining residues in crystal structures using geometric criteria and is able to predict the likelihood of a transmembrane helix being involved in pore formation. Results from testing this approach under stringent cross-validation indicate that prediction accuracy of 72% is possible, whereas a support vector regression model is able to predict the pore stoichiometry, i.e. the number of subunits required to form the pore in multimeric proteins, with 62% accuracy. This method provides a way to characterize pores in transmembrane proteins in the absence of crystal structures (19). HSPred HSPred (20) attempts to predict those residues in a protein–protein interface, which make the greatest contribution to the free energy of binding. Protein–protein interactions typically depend on just a handful of energetically critical ‘Hot Spot’ residues. HSPred is an SVM-based method, which attempts to identify such residues. The method takes each interface residue in turn, substitutes it for alanine and then predicts the change in the free energy of binding at the interface. These changes in free energy are then assessed by one of three SVM models; one trained against arginine to alanine mutations, one against glutamic acid to alanine mutations and one for all other substitutions. In predicting ‘Hot Spot’ residues, HSPred achieves a precision and recall of 61 and