PredictProtein--an open resource for online prediction of protein structural and functional features.
Yachdav G, Kloppmann E, Kajan L, Hecht M, Goldberg T, Hamp T, Hönigschmid P, Schafferhans A, Roos M, Bernhofer M, Richter L, Ashkenazy H, Punta M, Schlessinger A, Bromberg Y, Schneider R, Vriend G, Sander C, Ben-Tal N, Rost B. · 2014
Read PredictProtein--an open resource for online prediction of protein structural and functional features. by Yachdav G, Kloppmann E, Kajan L, Hecht M, Goldberg T, Hamp T, Hönigschmid P, Schafferhans A, Roos M, Bernhofer M, Richter L, Ashkenazy H, Punta M, Schlessinger A, Bromberg Y, Schneider R, Vriend G, Sander C, Ben-Tal N, Rost B. (2014) 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-pmc4086098.txt.
Introduction
Guy Yachdav, Burkhard Rost, and a large team presented PredictProtein, a web service that forecasts a protein's structural and functional features from its sequence alone. Their paper appeared in Nucleic Acids Research in 2014. The service had been running since 1992 and aims at experimentalists who are not bioinformatics specialists.
One query returns alignments plus predictions of shape features such as helices, membrane spans, and disordered regions. Added tools infer likely functions, cell location, binding sites for proteins and DNA or RNA, and the effect of single-letter changes. Results appear as text and as interactive figures meant to be read at a glance. Over twenty-two years the service grew from a few structure predictions into a broad annotation suite.
Catalog record
| CREATOR | Yachdav G, Kloppmann E, Kajan L, Hecht M, Goldberg T, Hamp T, Hönigschmid P, Schafferhans A, Roos M, Bernhofer M, Richter L, Ashkenazy H, Punta M, Schlessinger A, Bromberg Y, Schneider R, Vriend G, Sander C, Ben-Tal N, Rost B. |
|---|---|
| YEAR | 2014 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 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:
PredictProtein—an open resource for online prediction of protein structural and functional features By Guy Yachdav, Edda Kloppmann, Laszlo Kajan, Maximilian Hecht, Tatyana Goldberg, Tobias Hamp, Peter Hönigschmid, Andrea Schafferhans, Manfred Roos, Michael Bernhofer, Lothar Richter, Haim Ashkenazy Published in: Nucleic Acids Research DOI: 10.1093/nar/gku366 ABSTRACT PredictProtein is a meta-service for sequence analysis that has been predicting structural and functional features of proteins since 1992. Queried with a protein sequence it returns: multiple sequence alignments, predicted aspects of structure (secondary structure, solvent accessibility, transmembrane helices (TMSEG) and strands, coiled-coil regions, disulfide bonds and disordered regions) and function. The service incorporates analysis methods for the identification of functional regions (ConSurf), homology-based inference of Gene Ontology terms (metastudent), comprehensive subcellular localization prediction (LocTree3), protein–protein binding sites (ISIS2), protein–polynucleotide binding sites (SomeNA) and predictions of the effect of point mutations (non-synonymous SNPs) on protein function (SNAP2). Our goal has always been to develop a system optimized to meet the demands of experimentalists not highly experienced in bioinformatics. To this end, the PredictProtein results are presented as both text and a series of intuitive, interactive and visually appealing figures. The web server and sources are available at http://ppopen.rostlab.org. INTRODUCTION Molecular biology is moving into the high-throughput mode as the number of experiments needed to support a single hypothesis is rapidly growing. The line between experimental result and computational analysis is blurring; this also shifts what constitutes a reliable annotation. On top, the vast amount of life science data outpaces computer power. For example, less than 1% of the over 51 million sequences in UniProt (February 2014) (1) have some expert annotations in Swiss-Prot. This protein annotation gap widens every day (2). PredictProtein is one of the resources applicable to all proteins that contribute to closing this gap. The PredictProtein (PP) server is an automatic service that searches up-to-date public sequence databases, creates alignments, and predicts aspects of protein structure and function. In 1992, PredictProtein went online as one of the first Internet servers in molecular biology at the EMBL (Heidelberg, Germany). From 1999 to 2009, the server operated from Columbia University (New York, NY) and in 2009 it moved to the TUM (Munich, Germany). PredictProtein was one of the first services realizing state-of-the-art protein sequence analysis, and the prediction of structural and functional features in a single server. While many outstanding services (3) have expanded on some of those aspects, PredictProtein has remained one of the most comprehensive resources. The thousands of citations to PredictProtein and to our methods demonstrate the server's applicability and acceptance. Since 2009, for example, its website was visited more than one million times by about 80 000 unique visitors per year from 139 countries. Furthermore, over 500 000 sequences were submitted and processed by the service. About half of all submitted sequences were not in UniProt (1) at the time of submission. This suggests that the server's primary utility is in providing annotations for uncharacterized proteins. The following two central principles have guided the evolution of PredictProtein. Sustained quality with performance estimates. The performance of many tools is not sufficiently assessed and/or their performance does not sustain over time. Two decades of Critical Assessment of protein Structure Prediction (CASP)-like experiments (4,5) have demonstrated this repeatedly. PredictProtein went online with a method for the prediction of protein secondary structure (PHD (6)) and 22 years later the performance estimates for that method continue to be valid: a unique achievement.Ease of use. From the beginning we have aspired to make the use of our tools intuitive for all users. Unfortunately, the growth in size and scope continues to challenge the realization of this guiding principle. In 1992, the service provided alignments and secondary structure prediction; in 2014, it includes over 30 complex tools. Creating a unified, natural interface for these tools is challenging. Furthermore, we need to invest more resources to sustain the increasing usage as the data flood surges on. For example, most of our CPU goes into running PSI-BLAST (7). Since 2009, databases grew 10-fold whereas the CPU speed has only tripled, i.e. we need at least three times the number of CPUs we currently have to achieve the same ease in handling each job. METHODS PredictProtein incorporates over 30 tools Supplementary Table S1, Supporting Online Material provides a comprehensive list of all components. Database searches: sequences similar to the query are identified by standard, pairwise BLAST (8) and iterated PSI-BLAST (7) searches (9,10) against a non-redundant combination of PDB (11), Swiss-Prot (12) and TrEMBL (1). In addition, functional motifs are taken from PROSITE (13) and domains from Pfam (14). Prediction of structural features: predicted aspects of structure include PROFphd secondary structure and solvent accessibility (15,16), PROFtmb transmembrane strands (17), TMSEG transmembrane helices, COILS coiled-coil regions (18), DISULFIND disulfide bonds (19) and SEG low-complexity regions (20). Disordered regions are predicted by a set of tools: UCON (21), NORSnet (22), PROFbval (23,24) and Meta-Disorder (25). Prediction of functional features: predicted aspects include ConSurf annotations and visualizations of functionally important sites (26,27), protein mutability landscape analysis showing the effect of point mutations on protein function predicted by SNAP2 (28), Gene Ontology (GO) terms from metastudent (29), LocTree3 predictions of subcellular localization (30), protein–protein interaction sites (ISIS2) and protein–DNA, protein–RNA binding sites (SomeNA). Almost all prediction methods use evolutionary information obtained from PSI-BLAST searches; the more related protein sequences are found and the more divergent those are, the higher the gain in performance (10,15). However, none of the methods (with the exception of metastudent, see below) relies solely on profiles and the prediction without a profile is significantly better than random. For most prediction methods (e.g. LocTree3 and SNAP2) the prediction quality is estimated by a reliability score. In the following, we introduce some of the recent and upcoming additions since 2004 (31) in more detail. New: TMSEG transmembrane helix predictions TMSEG (Bernhofer, M. et al., in preparation) predicts alpha-helical transmembrane proteins, the position of transmembrane helices, and membrane topology. The method uses a novel segment-based neural network to refine the final prediction. TMSEG was developed and evaluated on 166 transmembrane proteins extracted from PDBTM (32) and OPM (33), and on 1441 proteins from the SignalP4.1 dataset (34). In our hands, TMSEG appears to complement and improve over the best existing methods (e.g. PolyPhobius (35) and Memsat3 (36)) predicting all membrane helices correctly for about 60% of all proteins. The method correctly identifies 98% of all transmembrane proteins with a false positive rate of less than 2%. New: SNAP2 predict effect of mutations upon function SNAP2 predicts the effect of single amino acid substitutions on protein function (37). It improves over its predecessor SNAP (38) by using additional coarse-grained features that better classify samples with unclear evidence. With a two-state accuracy of 83% and an AUC of 0.91, SNAP2 performs on par or better than other state-of-the-art methods on human variants while significantly outperforming these methods for other organisms. SNAP2 is the only available method predicting the effect of point mutations even without alignment information (if fewer than 10 related proteins are found, a specific method is applied with an expected accuracy of ∼70% instead of 83%). For each protein we also predict the entire protein mutability landscape (28,39), i.e. the functional effect of