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
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.
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
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
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
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
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.
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
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.