SBC logo KTH logo KI logo SU logo SSF logo

ProQ - Protein Quality Predictor

ProQ is a neural network based predictor that based on a number of structural features predicts the quality of a protein model. ProQ is optimized to find correct models in contrast to other methods which are optimized to find native structures. Two quality measures are predicited LGscore and MaxSub.

LGscore is -log of a P-value and MaxSub ranges from 0-1, were 0 is insignificant and 1 very significant.

Different ranges of quality:
LGscore > 1.5
MaxSub  > 0.1
LGscore > 3
MaxSub  > 0.5
Very good
LGscore > 5
MaxSub  > 0.8

ProQ web server

Stand alone version of ProQ

A stand alone version of ProQ for linux can be download here. Before running the program please see the instructions in the README.

Build all-atom models

You can build all-atom models from alignment on a known structure here. To use it you need an academic license of MODELLER. The necessary key can be obtained here. The model is automatically evalutated using ProQ. The scores are supplied in the header of the model as:


Additional tables to the ProQ paper.

Results for the different methods for each protein on known Decoy Sets and on models created from LiveBench-2. All atom models from LiveBench-2 can be downloaded here (205MB). Each file has the MaxSub, LGscore, RMSD and when ever possible also the Pcons score in the header.

Known Decoy Sets LiveBench-2
ProQ-combined ProQ-combined
Errat Errat
ProsaII ProsaII
Verify3D Verify3D

For each method the following tables is presented:

Protein decoys rank MX LG RMSD Z Z-nat
1ctf 630 1 1.00 4.11 0.00 2.7 5.1
1r69 675 2 0.82 2.80 1.66 3.4 6.2
1sn3 660 1 1.00 3.99 0.00 1.9 4.6
2cro 674 1 1.00 3.99 0.00 2.0 4.8

Protein name of target protein
decoys number of decoys in the set
rank rank of the native structure
MX MaxSub for the highest ranked model
LG LGscore for the highest ranked model
RMSD RMSD for the highest ranked model
Z Z-score for seperating incorrect and correct models
Z-nat Z-score for the native structure