Critical Assessment of Fully Automated Structure Prediction


CAFASP3 Registered Servers*

Server Type Short Description Server Developer / Person in Charge Contact e-mail Address CASP5 Server Registration
ESYP HM ESyPred3D exploits a new alignment strategy using neural networks. Alignments obtained by combining, weighting and screening several multiple alignment programs. Final structure built using MODELLER Christophe Lambert Christophe.lambert@fundp.ac.be ESyPred3D
MODZ HM Modzinger Z first searches the PDB using Psi-blast. Next, a multiple structure alignment is derived, to which the target is aligned. The fragments with most conserved sequence are used to build an initial model. The rest is built using fragments from a database Shoshana Wodak mz@ucmb.ulb.ac.be Not Registered
MODW HM ModWeb is a server for calculating homology models for individual domains for the target sequence. It can also a structure to calculate models of sequences homologous to it Andrej Sali sali@rockvax.rockefeller.edu Not Registered
MOUL HM MOULDER is a homology modeling server Andrej Sali sali@rockvax.rockefeller.edu Not Registered
JIGS HM 3D-Jigsaw looks for homologous templates in a sequence databases (PFAM+PDB+nr) and splits the query sequence into domains. If good templates are found, the best covered domain is then modelled using a maximum of 2 Paul Bates paul.bates@cancer.org.uk 3D-JIGSAW
FAMS HM FAMS is a homology modeling server Hideaki Umeyama iwadatem@pharm.kitasato-u.ac.jp FAMS
FAMD HM FAMSD is a new version of FAMS Hideaki Umeyama iwadatem@pharm.kitasato-u.ac.jp FAMSD
PRCM HM PROTINFO-CM is the HM part of the PROTINFO server. The CM component does a sequence-only search to build a multiple alignment. Initial models are then built from templates and scored. Loops and side chains are built on the best models using a frozen approximation. Ram Samudrala ram@compbio.washington.edu PROTINFO-CM
TSUN HM TSUNAMI uses a BLAST-like algorithm enhanced with statistical potentials to search a structural database. A number of recognized homologous fragments is used to construct a final model Piotr Rotkiewicz piro@chem.uw.edu.pl Tsunami
ALAX HM Alax is a new sequence alignment method based on solvent accessibilities of template structures. Final structures are built using the homology modeling program FAMS Mitiko Go alax@biol1.bio.nagoya-u.ac.jp ALAX
PCOB HM/FR Pcomb uses a combination of psi-blast, profile-profile and predictions based alignments methods to create template alignments for homology. modelling Arne Elofsson arne@sbc.su.se Pcomb
PRFR FR PROTINFO-FR is the FR part of the PROTINFO server. The RAMP software uses a de novo method for generating a particular fold. Now we use CM methods to identify tenuous sequence relationships. Once a template has been chosen, it is used as a beacon to guide our simulations and a large number of structures are generated and the best scoring is selected Ram Samudrala ram@compbio.washington.edu PROTINFO-FR
RAPT FR RAPTOR uses a protein threading technique to do fold recognition via minmizing an energy function consisting of mutation, singleton, pairwise, secondary structure items. The problem is formulated as a large scale integer programming problem. Finally SVM technique is used to assess the alignment reliability Ming Li j3xu@math.uwaterloo.ca RAPTOR
MPAL FR MPALIGN combines various alignment methods. PSI-blast for easy targets. For the others, DP based sequence-to-profile alignment, using profiles from (i) PSI-search from each representative sequence, (ii) multiple profiles from sequences in the same fold, (iii) combination of (i) and profiles from structurally similar fragments. Alignment scores against many sequences are considered for selecting a template. Tatsuya Akutsu takutsu@kuicr.kyoto-u.ac.jp MPALIGN
PFIN ? PROTFINDER aligns the sequences to structures in the PDB_SELECT set. Scores result from an effective contact energy plus gap penalty terms. Near optimal alignments constructed through the Build-up algorithm (Park and Levitt, 1994). Dr Ugo Bastolla bastollau@inta.es Protfinder
CMAP ? CMap23Dpro is a contact 2d and 3d structure prediction server Gianluca Pollastri gpollast@ics.uci.edu CMap23Dpro
ARBY FR ARBY is a compute intensive method from Fraunhofer-Institute for Algorithms and Scientific Computing (SCAI) Ingolf Sommer ingolf.sommer@scai.fhg.de arby-scai
PROS FR PROSPECT uses threading based methodology to carry out protein fold recognition Ying Xu x9m@ornl.gov ORNL-PROSPECT
FFAS FR FFAS compares sequence profiles with each other. Profiles are generated for protein families in a different way than in PSI-Blast but PSI-Blast is used to collect the proteins of a family. The official version from the Burnham institute is used here Leszek Rychlewski leszek@bioinfo.pl FFAS
FFA3 FR FFAS03 is based on profile-profile comparison algorithm, similar to algorithm used in FFAS and Basic family of methods. Most important differences are in calculation of profile-profile matching and in empirical evaluation of significance. FFAS03 was created at Godzik Lab at The Burnham Institute Lukasz Jaroszewki lukasz@ljcrf.edu FFAS03
FOR1 FR FORTE1 is a profile-based fold recognition program Kentaro Tomii k-tomii@aist.go.jp FORTE1
LOOP FR LOOPP (Learning, Observing and Outputting Protein Patterns) aligns sequence to sequence, sequence to structure, and structure to structure. It further enables the optimization of potentials and scoring functions for the above applications Ron Elber loopp@tc.cornell.edu Ron-Elber
ORFs FR ORFeus is based on the alignment of two profiles, which include sequence information from the family and predicted secondary structure. ORFb is used to search for the initial family Leszek Rychlewski leszek@bioinfo.pl BioInfo.PL-ORFeus
BasB FR BASIC bilaterally amplified sequence information comparison. Version B is very conservative. It uses only 3 PSI-Blast iterations Leszek Rychlewski leszek@bioinfo.pl BioInfo.PL-BasicB
BasC FR BASIC bilaterally amplified sequence information comparison. Version C is very aggressive. It uses 7 PSI-Blast iterations and RPS-Blast

Leszek Rychlewski leszek@bioinfo.pl BioInfo.PL-BasicC
ST99 FR Sam-T99 builds a multiple alignment by iterated search using HMMs to predict secondary structure and to build an HMM used for searching the PDB. Also, a library of HMMs is used to score the target sequence Kevin Karplus karplus@soe.ucsc.edu SAM-T99
ST02 FR Sam-T02 new version of the Sam-T server Kevin Karplus karplus@soe.ucsc.edu SAM-T02-server
SFAM FR SUPERFAMILY is a library of HMMs based on SCOP. Aimed at assigning SCOP domains to sequences at the superfamily level. Uses SAM software; updated with each SCOP release Julian Gough jgough@mrc-lmb.cam.ac.uk SUPERFAMILY
SFPP FR SUPFAM_PP is the next generation of SUPERFAMILY. Designed for whole genome analysis, it uses an HMM-based profile-profile comparison method Julian Gough jgough@mrc-lmb.cam.ac.uk SUPFAM_PP
3DPS FR 3D-PSSM is based on a threading approach using 1D and 3D profiles coupled with secondary structure and solvation potential. Updates its fold library every week Lawrence Kelley L.A.kelley@ic.ac.uk 3D-PSSM
GETH FR GenTHREADER combines various methods including sequence alignment with structure based scoring functions and it uses a neural network based jury system to calculate the final score for the alignment David Jones David.Jones@cs.ucl.ac.uk GenTHREADER
MGTH FR mGenTHREADER is an enhanced version of GETH. It takes as input a PSI-BLAST profile calculated for the target sequence David Jones David.Jones@cs.ucl.ac.uk mGenTHREADER
FUG2 FR FUGUE uses environment-specific substitution tables derived from the HOMSTRAD database, along with structure-dependent gap penalties, to construct profiles. FUGUE2 is an enhanced version of FUGUE, where the structural profiles are enriched by homologous sequences. Kenji Mizuguchi kenji@cryst.bioc.cam.ac.uk FUGUE2
FUG3 FR FUGUE3 is a new experimental version of FUGUE, which uses a slightly different profile library. Kenji Mizuguchi kenji@cryst.bioc.cam.ac.uk FUGUE3
INBG FR INBGU combines 5 methods which exploit sequence and structure information in different ways to produce a consensus prediction of the 5. Uses predicted vs observed secondary structure and sequence profiles. Daniel Fischer dfischer@cs.bgu.ac.il INBGU
SHGU FR ShotGun-INBGU uses the ShotGun method utilizing the INBGU components. It assembles hybrid models from INBGU. Daniel Fischer dfischer@cs.bgu.ac.il 3D-SHOTGUN-INBGU
RPFD FR RPFOLD is a hybrid method, using various sequence similarity techniques (like SSEARCH & PSI-BLAST) and secondary structure information Raghava raghava@imtech.res.in RPFOLD/ Raghava-Garjendra
BLAS FR Simple Blast to search for trivial homologs in the PDB. Runs locally at bioinfo.pl Leszek Rychlewski leszek@bioinfo.pl Not Registered
PDBb FR PDB-Blast is based on the PSI-Blast program. Before the fifth iteration the sequence profile is saved and used as query against the PDB sequences. Runs locally at bioinfo.pl Leszek Rychlewski leszek@bioinfo.pl BioInfo.PL-PDB-Blast
ORFb FR ORF-Blast uses the Blast tools for iterative search for homologs in PDB (as PDB-Blast). The initial family of homologs is built using RPS-Blast. It is expanded using 5 iterations of PSI-Blast on nr (nr70) Leszek Rychlewski leszek@bioinfo.pl BioInfo.PL-ORFblast
PRAB AI PROTINFO-AB is the ab initio part of PROTINFO. Shorter sequences without related templates, are modelled using de novo methods. Ram Samudrala ram@compbio.washington.edu PROTINFO-AB

In addition, and in a different category, are meta-predictors which generate predictions utilizing models produced by other services.

Meta-Predictor Type Short Description Server Developer / Person in Charge Contact e-mail Address CASP5 Server Registration
CONS ? CONSENSUS predicts only the high confidence structural alignment of a target using a hybrid of other methods available in the literature Carlos Camacho ccamacho@bu.edu Camacho-Carlos
RBTA HM/AI ROBETTA provides both ab initio and comparative models of protein domains. It uses the ROSETTA fragment insertion method. Comparative models built from structures detected by PDB-BLAST or Pcons2 and aligned by the K*SYNC alignment method. David Baker dabaker@u.washington.edu BAKER-ROBETTA
HMMS AI HMMSTR, I_SITES and ROSETTA used to produce ab initio predictions using the ROSETTA Fragment Insertion Monte carlo method. The fragment moveset is made by I_SITES. Secondary structure and local structure are predicted by HMMSTR Chris Bystroff bystrc@rpi.edu I-sites/Bystroff
CNBP FR CNBPred combines results from different threading servers, among them and with other information Alfonso Valencia osvaldog@cnb.uam.es CNBpred
LBLL FR LIBELLULA obtains predictions from samt99 and 3dpssm, then a NN REevaluates with additional evolutionary information Alfonso Valencia osvaldog@cnb.uam.es LIBELLULA
PCO2 FR Pcons2 (updated Pcons) is a consensus server that utilizes the predictions from PDB-Blast, FFAS, 3D-PSSM, GenTHREADER, mGenTHREADER, INBGU, Sam-T99 and FUGUE. It selects one of the models returned by these services Arne Elofsson arne@sbc.su.se Pcons2
PCO3 FR Pcons3 (new Pcons) is a consensus server that utilizes the predictions from PDB-Blast, FFAS, 3D-PSSM, GenTHREADER, mGenTHREADER, INBGU, Sam-T99, FUGUE2, SHotGun-INBGU and ORFeus. It selects one of the models returned by these services Arne Elofsson arne@sbc.su.se Pcons3
PMOD HM/FR Pmodeller is based on Pcons2. It provides models optimized in the last step using the Modeller program Arne Elofsson arne@sbc.su.se Pmodel
PMO3 HM/FR Pmodeller3 is based on Pcons3. It provides models optimized in the last step using the Modeller program Arne Elofsson arne@sbc.su.se Pmodel3
3DS3 FR ShotGun on 3 is a meta-predictor using the ShotGun method applied to the results of FFAS, 3D-PSSM and INBGU. It assembles hybrid models from models of the 3 servers. Daniel Fischer dfischer@cs.bgu.ac.il 3D-SHOTGUN-3DS3
3DSN FR 3DSN returns the same result as 3DS3, but without the assembly Daniel Fischer dfischer@cs.bgu.ac.il 3D-SHOTGUN-3DSN
3DS5 FR ShotGun on 5 is a meta-predictor, like 3DS3, but using the results of FFAS, 3D-PSSM, SHGU, FUGUE and GenTHREADER. Daniel Fischer dfischer@cs.bgu.ac.il 3D-SHOTGUN-3DS5




*Servers for Secondary Structure and Contacts Prediction are not yet listed.

several