Assessing interface accuracy in macromolecular complexes.

Area of Science:

• Structural bioinformatics
• Computational biology
• Biophysics

Background:

• Accurate 3D structure prediction of macromolecular complexes is crucial for understanding cellular functions.
• Evaluating the quality of these in silico models remains a significant challenge in bioinformatics.
• Existing methods assess models from multiple perspectives, including component structure and complex arrangement.

Purpose of the Study:

• To introduce and validate new measures for assessing the quality of predicted 3D macromolecular complex structures.
• To specifically quantify intermolecular interactions and evaluate interfaces in multichain complexes.
• To provide reliable scoring and ranking of prediction models.

Main Methods:

• Introduction of Intermolecular Interaction Network Fidelity (I-INF), a normalized similarity measure for intermolecular interactions in multichain complexes.
• Adaptation of I-INF from a score used in the RNA field, focusing on interchain interaction sites.
• Implementation of the F1 measure to assess interfaces in macromolecular assemblies.
• Testing on diverse complexes including RNA-protein, DNA-DNA, RNA-RNA, and protein-protein decoys.

Main Results:

• I-INF and F1 measures provide clear, intuitive, and reliable scores for evaluating predicted 3D models.
• The implemented measures enable straightforward ranking of prediction quality.
• Successful validation across various types of macromolecular complexes, including 72 RNA-protein decoys.

Conclusions:

• I-INF and F1 offer robust metrics for assessing the accuracy of predicted macromolecular complex structures.
• These measures enhance the evaluation framework for computational models of molecular assemblies.
• A publicly available tool for computing I-INF and F1 facilitates integration and large-scale analysis.