A machine learning-based method for protein global model quality assessment

Abstract

Model quality assessment is an important task in protein structure prediction, which has been recently identified as one of the bottlenecks limiting the quality and usefulness of protein structure prediction methods. A new prediction category that evaluates the quality of protein models has been implemented since CASP7. In this study, a machine learning-based method for protein global model quality assessment (GMQA) is presented and the web server can be assessed at http://www.iipl.fudan.edu.cn/gmqa/index.html. The GMQA method takes a protein model as input and outputs the predicted MaxSub score that indicates the absolute quality of the protein model. The proposed method extracts the structural features from the 3D coordinates and assigns an absolute quality score to a model by support vector regression. Three types of features are extracted, including secondary structure, relative solvent accessibility, and contact. The closed test is performed on the CASP7 data set using cross validation. An open testing is performed on the LKF data set. In both tests, good correlations between the predicted and true scores are observed. Furthermore, our method is able to discriminate the native or near-native structures from a set of decoys. We also demonstrate that the GMQA method outperforms two existing methods, i.e. ProQ and Victor/FRST. Our results show that GMQA is a useful tool for model quality assessment and ranking.

Keywords:

• protein structure prediction
• model quality assessment
• support vector regression

Acknowledgements

This work was supported by the National Natural Science Foundation of China (NSFC) (No. 60873040 and No. 60873070), Shanghai Leading Academic Discipline Project (No. B114), and the China Postdoctoral Science Foundation funded project (No. 20080440572).

Notes

1. http://predictioncenter.org.

2. http://titan.princeton.edu/decoys/.