Abstract/Summary

Proteins play a crucial role in the biological machinery of living organisms, with their structures dictating functions essential for life processes. Disruptions in protein function lead to diseases. Therefore, knowledge of proteins is vital for biomedical sciences and biotechnology. Protein structures are experimentally determined by NMR or X-ray crystallography, but computational methods have gained prominence due to their speed and accuracy. Recent advances in protein structure prediction provide high-accuracy 3D models, challenging quality estimation methods. Predicting residue contacts can be useful in obtaining significant information that may be used to improve the performance of quality estimation methods. Contact prediction methods have evolved, utilising diverse protein databases and approaches to enhance 3D protein model accuracy. However, challenges persist in modelling certain targets. This study proposes consensus approaches, combining data from deep learning-based contact prediction methods from CASP13 and CASP14, leading to measurable advancements in accuracy. We then investigated the role of consensus contact prediction in improving the performance of ModFOLD9 using the CDA score. The experiment expanded to integrate various quality scores derived from the pure-single model and quasi-single model methods to further enhance ModFOLD9's accuracy. The consensus algorithms and contact prediction improved ModFOLD9's local quality estimations for tertiary structure models. This strategy was extended to enhance the IntFOLD7 and ModFOLDdockS servers. We analysed the performance of the improved servers using two gold-standard blind experiments: CAMEO and CASP15. The evaluation of these servers validated their improved performance and highlighted the impact of contact prediction on enhancing both local tertiary structure model quality estimations and quaternary structure model quality estimates for interface residues. Overall, our study demonstrated the importance of contact prediction in improving the performance of model quality estimation tools in the field of protein structure prediction.