Emerging Frontiers in Conformational Exploration of Disordered Proteins: Integrating Autoencoder and Molecular...

Intrinsically disordered proteins (IDPs) are closely associated with a number of neurodegenerative diseases, such as Alzheimer's disease and Parkinson's disease. Due to the highly dynamic nature of IDPs, their structural determination and conformational exploration pose significant challenges for both experimental and computational research. Recently, the integration of machine learning with molecular dynamics (MD) simulations has emerged as a promising methodology for efficiently exploring the conformation spaces of IDPs. In this viewpoint, we briefly review recently developed autoencoder-based models designed to enhance the conformational exploration of IDPs through embedding and latent sampling. We highlight the capability of autoencoders in expanding the conformations sampled by MD simulations and discuss their limitations due to the non-Gaussian latent space distribution and the limited conformational diversity of training conformations. Potential strategies to overcome these limitations are also discussed.