Abstract
Artificial intelligence (AI)-based 3D protein modelling software have revolutionized structural biology, often predicting protein structures with unprecedented confidence. However, to get the most out of AI, it is advisable to consider the information provided by a set of 'best' models with similar reliability scores, rather than just the top-ranked model, and to compare the results of different AI predictors. Furthermore, the 'highest scoring' solution is not always the 'best' one, especially in the case of oligomeric models where it may be crucial to introduce experimental knowledge into their selection. Here, we present the FoldScript web server, which has been developed to address these issues. It can be used to gather and synthesize primary to quaternary structure information from a set of up to 25 AI-generated models. The analysis is presented in the form of a comprehensible flat figure and mosaics of 3D structures, allowing efficient visual inspection of structural differences. Another module identifies models with known intermolecular contacts. FoldScript is therefore a decision support tool designed to help a large community, whether or not they are structural biologists, to rationally select the most accurate model in an AI-driven modelling process. It is freely available without login requirement at https://foldscript.ibcp.fr.