Patrick Senet
We can now predict protein structures with remarkable accuracy from their amino acid sequences using deep learning. Yet proteins are not static objects—they explore rich and dynamic conformational ensembles. How temperature and mutations reshape these ensembles remains largely unknown.
The traditional two-state view of protein folding—folded versus unfolded—misses much of this complexity. Capturing folding pathways, misfolded states, and intrinsically disordered regions, all of which are closely linked to disease, remains a major challenge. Here, we take a different perspective. By representing molecular dynamics ensembles as graphs and by extracting global and local descriptors, we uncover residue-level order–disorder transitions during folding.
We validate this approach against NMR data for the model protein gpW, and we show how a single disease-related mutation reshapes the conformational landscape of α-synuclein.