Generating Free-energy Landscapes and Mapping Conformational Transition Pathways from cryo-EM Images using a Deep Isometric Autoencoder

Biological functions of macromolecules are regulated by their three-dimensional structures and dynamic behaviors. Cryo-electron microscopy (cryo-EM) is increasingly gaining attention for determining novel static structures with large molecular sizes and complex structural assemblies. However, investigating the dynamic behaviors of these structures on free-energy landscapes is substantially difficult. Herein, we developed a deep-learning-based method, PaStEL with cryoTWIN, to uncover thermodynamic behaviors recorded in cryo-EM data. CryoTWIN trained with cryo-EM particle images builds an isometric latent space for a conformational space of the free-energy landscape, allowing PaStEL to generate the energy-based conformational transition pathways among the heterogeneous structural states. The theoretical discoveries and the simulation-based experiments guaranteed our approach. Furthermore, using experimental data, parallel assembly pathways in 50S-ribosome and the thermodynamic basis of the enhanced infectivity of SARS-CoV-2 spike protein were elucidated without prior biological knowledge, thereby highlighting the impact of our method in structural biology researches. Thus, PaStEL with cryoTWIN had a potential to advance biological foundation by depicting whole pictures of the dynamic behaviors.