QMEANDisCo-distance constraints applied on model quality estimation.

Area of Science:

• Structural bioinformatics
• Computational biology
• Protein modeling

Background:

• Accurate estimation of 3D protein model quality is essential for determining model utility.
• Existing methods include single-model approaches and consensus methods requiring model ensembles.
• There is a need for methods that combine the strengths of both approaches.

Purpose of the Study:

• To extend the QMEAN score by incorporating a consensus-based distance constraint (DisCo) for improved 3D protein model quality estimation.
• To develop a composite score, QMEANDisCo, that integrates single-model and consensus-based features.
• To evaluate the performance of QMEANDisCo and compare it with existing methods.

Main Methods:

• Introduction of the DisCo score, leveraging distance distributions from homologous experimental structures.
• Training feed-forward neural networks to adaptively weigh DisCo and classical QMEAN parameters.
• Development of the composite score QMEANDisCo.

Main Results:

• QMEANDisCo effectively combines the accuracy of consensus methods with the broad applicability of single-model approaches.
• The study demonstrates that CASP models are suboptimal for training predictive quality estimation methods.
• QMEANDisCo consistently ranks among top performers in CAMEO and CASP13 benchmarking with low response times.

Conclusions:

• QMEANDisCo represents a significant advancement in computational protein structure quality assessment.
• The method offers a robust and efficient tool for evaluating the quality of predicted protein models.
• QMEANDisCo is available as a web server and its source code can be downloaded.