MQAPRank: improved global protein model quality assessment by learning-to-rank

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Protein structure prediction has advanced, generating numerous models for a given sequence.
• Accurate assessment of predicted protein model quality is crucial for successful structure prediction.
• Existing quality assessment methods (single, quasi-single, clustering) have limitations, leaving accurate assessment an open problem.

Purpose of the Study:

• To develop a novel global protein model quality assessment program.
• To improve the accuracy and efficiency of protein model quality assessment.

Main Methods:

• Developed MQAPRank, a global protein model quality assessment program utilizing a learning-to-rank approach.
• MQAPRank initially ranks decoy models by quality using a learning-to-rank algorithm.
• Subsequently, it uses the top five ranked models as references to predict the quality of other models based on average GDT_TS scores.

Main Results:

• MQAPRank demonstrated superior performance compared to leading protein model quality assessment methods on CASP11 and 3DRobot datasets.
• The program achieved state-of-the-art results in the CASP12 competition under the group name FDUBio.

Conclusions:

• MQAPRank offers a powerful and user-friendly tool for protein model quality assessment.
• The tool achieves state-of-the-art performance, benefiting protein structure prediction and model quality assessment applications.