GMQ: Prediction of Local Quality of Protein Structure Models Considering Spatial Neighbors in Graphical Models
GMQ is a protein structure quality aseessment program using graphical conditional random field. It uses 25 features as a input to describe residue character. To see the detailed descriptions, please read the our paper cited below:
W.-H. Shin, X. Kang, J. Zhang, and D. Kihara (2016), Sci. Rep. under revision
- Obtaining CrfAny: For the CRFs, we use the open source from http://sourceforge.net/projects/crfany/. Please download the source code and compile it before run GMQ.
- Downloding GMQ package: To run GMQ, GMQ package should be downloaded. After decompressing the file, there are 'example' directory containing recent CASP target, 'README.txt' gives a brief explanation of input feature files, 'gmq.py' allowing users to train and test by themselves, and 'test_run' for testing CrfAny can work properly or not.
To run the files in 'test_run', execute the python script as follows:
python gmq.py ca_cutoff neighbor_cutoff max_neighbor weight_hh weight_bb weigth_hb/bh weight_hc/che.g.) python gmq.py 5 4.5 3 0.8 0.8 0.9 0.1
To train and test with your own protein models, the files should be prepared as in 'test_run'. You need to put the train and test file in the “data” folder, residue-residue distance file in “residue_pair” folder, and secondary structure file in “2ndstructure” folder. The result will be in the folder “result”. The model is the trained model and the output is the prediction for the test data.
- Evaluating CASP target example: One of recent CASP12 target, T0948, is prepared in example directory of GMQ package. In the README.txt, a brief explanation of running GMQ is described.
Contact: dkihara@purdue.edu