Abstract
The optimization of diagnostic devices such as biosensors often requires understanding the molecular details of the interaction between capture and target biomolecules. This can be experimentally obtained by cryo-electron microscopy, the preferred method for the analysis of large protein complexes, while NMR and x-ray crystallography are effective for determining the structure of complexes formed by relatively small molecules. Nevertheless, all these approaches are demanding in terms of time and resources and, therefore, we explored the possibility to reduce the experimental load by compensating with in silico modelling. Here we demonstrate that an accurate prediction of the binding mode between a nanobody and its target pea ascorbate peroxidase, an oxidative stress biomarker in plants, can be obtained by combining cross-linking mass spectrometry, hydrogen-deuterium exchange coupled to mass spectrometry and in silico modelling. Such model allowed to precisely design negative mutants that confirmed its accuracy. In conclusion, this study shows that an unconstrained prediction based on deep learning models is still not sufficiently reliable for new targets and difficult-to-model biomolecule classes such as nanobodies, while an experimental-guided approach can provide valuable structural information for lead optimization campaigns of such reagents.