You don't have javascript enabled. The web-server may behave improperly.

New query
Download
References
News
Server status
Example results
ProQ2
Help

Your recent jobs:

Queued    0
Running    0
Finished   0
Failed    0

© Arne Elofsson

SciLifeLab Logotype nbis Logotype
SeRC Logotype Stockholm University logotype EGI logotype

Help



1. Summary


ProQ3 is a Model Quality Assessment (MQA) program that predicts the quality of model structures based on a single model. It employs both the structural information derived from the three-dimensional (3D) structure of the model to be predicted and the evolutionary information derived from the amino acid sequence of this model. ProQ3 is based on a combination of our previously developed ProQ2 and two newly developed methods ProQCenFA and ProQRosCen, where ProQCenFA uses full-atom model and ProQRosCen uses centroid model. Our benchmark based on the CASP11 dataset and CAMEO dataset shows that ProQ3 outperforms the best previously developed MQA method, ProQ2. ProQ3 server outputs results for ProQ3, ProQCenFA, ProQRosCen and ProQ2 methods.

The original ProQ2/ProQ3 were trained using Support Vector Machines (SVM) with a linear kernel. Recently, we have retrained these methods using the Deep Learning approach, implemented in Python with the Keras library using the Theano backend. This approach has significantly improved the performance of our methods. The Deep Learning versions of ProQ2/ProQ3, i.e. ProQ2D/ProQ3D have been implemented in both this web-server and the stand-alone program on github. To use ProQ2D/ProQ3D on the web-server, you need to check the option "Using deep learning". For the stand-alone program, you need to set the option "-deep" to "yes".

ProQ3/ProQ3D is supported by the EGI FedCloud with the VO VO.NBIS.SE for computational resource.



2. Usage


Input to the server are structure models in the PDB format. You can input the model structures by either paste your data in the first text area or upload a text file containing your data. You may upload multiple models (up to 5) in one model file. Multiple models should be recognized by the tag "MODEL N" and "ENDMDL" (see format description below).

If all your models are based on the same target sequence, you are recommended to input your target sequence in FASTA format. You can input your target sequence by either paste your sequence in the second text area or upload a text file containing your sequence under this text area. In that case, profiles are calculated only once for this target sequence, and the global scores are normalized by the length of the target sequence as well.

To generate an example input, just click the button in the panel New query and then click .

Format of an input model with multiple models
  MODEL 0
  ATOM      1  N   ILE    25      20.054  -1.774  13.539  1.00 13.86
  ATOM      2  CA  ILE    25      19.853  -1.767  12.074  1.00 13.86
  ATOM      3  C   ILE    25      19.368  -0.432  11.629  1.00 13.86
  ...                                                               
  ENDMDL                                                            
  MODEL 1                                                           
  ATOM      1  N   THR     2       6.845  39.808 112.651  1.00  7.56
  ATOM      2  CA  THR     2       7.837  40.283 113.642  1.00  7.56
  ATOM      3  C   THR     2       8.964  39.313 113.764  1.00  7.56
  ATOM      4  O   THR     2       8.919  38.204 113.232  1.00  7.56
  ATOM      5  1H  THR     2       6.111  40.299 112.478  1.00  7.56
  ENDMDL                                                            
            



3. Output


The server outputs both the global and local scores of the models to be predicted. ProQ3 server outputs results for ProQ3, ProQCenFA, ProQRosCen and ProQ2 methods.

If the target sequence is not given nor the target length is provided, the global scores are normalized by the length of the model sequence. Otherwise, the global scores are normalized by the target length.

A link to the dumped text file with both global scores and local scores is also provided in the result page given the successful prediction.

Additionally, a zipped folder with all intermediate result files is also downloadable.

An example of the predicted result can be found here



4. References


ProQ3: [Please cite this paper if you find ProQ3 useful in your research]
ProQ3: Improved model quality assessments using Rosetta energy terms. Karolis Uziela, Nanjiang Shu, Björn Wallner and Arne Elofsson. Scientific Reports, 33509. [PubMed]

ProQ3D: [Please cite this paper if you find ProQ3D useful in your research]
ProQ3D: Improved model quality assessments using Deep Learning. Karolis Uziela, David Menéndez Hurtado, Nanjiang Shu, Björn Wallner and Arne Elofsson. Bioinformatics. 2017 May 15;33(10):1578-1580 [PubMed]

Others:
Improved protein model quality assessments by changing the target function. Karolis Uziela, David Menéndez Hurtado, Nanjiang Shu, Björn Wallner and Arne Elofsson. Proteins. 2018 Jun;86(6):654-663. [PubMed]


5. Contact


Arne Elofsson group

Department for Biochemistry and Biophysics
The Arrhenius Laboratories for Natural Sciences
Stockholm University
SE-106 91 Stockholm, Sweden

Science for Life Laboratory
Box 1031, 17121 Solna, Sweden

E-mail:   arne@bioinfo.se
Phone:   (+46)-8-16 4672
Fax:   (+46)-8-15 3679