Estimation of model accuracy in CASP15 using the ModFOLDdock server.

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• CASP15 (Critical Assessment of protein Structure Prediction) showed increased focus on multimeric protein modeling, necessitating robust quality assessment (QA) for quaternary structures.
• Objective QA is crucial for evaluating the accuracy of protein assembly models.

Purpose of the Study:

• To develop and evaluate ModFOLDdock, a multimeric model QA server, for the CASP15 competition.
• To assess the performance of three ModFOLDdock variants in predicting the accuracy of protein multimeric models.

Main Methods:

• ModFOLDdock integrates single-model, clustering, and deep learning approaches to provide consensus QA scores.
• Three variants (standard, ModFOLDdockR for ranking, ModFOLDdockS for individual scoring) were optimized for CASP15.
• Performance was evaluated against CASP observed scores (oligo-lDDT) for homomeric and heteromeric models.

Main Results:

• All ModFOLDdock variants achieved Pearson correlation coefficients > 0.70 with observed scores.
• ModFOLDdock variants consistently ranked among the top two methods across all Estimation of Model Accuracy (EMA) categories.
• Specific high rankings were achieved for global fold prediction, interface quality prediction, and individual residue confidence scores.

Conclusions:

• ModFOLDdock variants provide reliable quality assessment for multimeric protein models.
• The server demonstrates competitive performance in predicting various aspects of model accuracy, including global structure and interface quality.
• ModFOLDdock is a valuable tool for the structural biology community, available as a web server and Docker package.