Assessing interface accuracy in macromolecular complexes.

Accurately predicting the 3D structures of macromolecular complexes is becoming increasingly important for understanding their cellular functions. At the same time, reliably assessing prediction quality remains a significant challenge in bioinformatics. To address this, various methods analyze and evaluate in silico models from multiple perspectives, accounting for both the reconstructed components' structures and their arrangement within the complex. In this work, we introduce Intermolecular Interaction Network Fidelity (I-INF), a normalized similarity measure that quantifies intermolecular interactions in multichain complexes. Adapted from a well-established score in the RNA field, I-INF provides a clear and intuitive way to evaluate the predicted 3D models against a reference structure, with a specific focus on interchain interaction sites. Additionally, we implement the F1 measure to assess interfaces in macromolecular assemblies, further enriching the evaluation framework. Tested on 72 RNA-protein decoys, as well as exemplary DNA-DNA, RNA-RNA, and protein-protein complexes, these measures deliver reliable scores and enable straightforward ranking of predictions. The tool for computing I-INF and F1 is publicly available on Zenodo, facilitating large-scale analysis and integration with other computational systems.