A Deep Learning-Driven Sampling Technique to Explore\nthe Phase Space of an RNA Stem-Loop

The\nfolding and unfolding of RNA stem-loops are critical biological\nprocesses; however, their computational studies are often hampered\nby the ruggedness of their folding landscape, necessitating long simulation\ntimes at the atomistic scale. Here, we adapted DeepDriveMD (DDMD),\nan advanced deep learning-driven sampling technique originally developed\nfor protein folding, to address the challenges of RNA stem-loop folding.\nAlthough tempering- and order parameter-based techniques are commonly\nused for similar rare-event problems, the computational costs or the\nneed for a priori knowledge about the system often present a challenge\nin their effective use. DDMD overcomes these challenges by adaptively\nlearning from an ensemble of running MD simulations using generic\ncontact maps as the raw input. DeepDriveMD enables on-the-fly learning\nof a low-dimensional latent representation and guides the simulation\ntoward the undersampled regions while optimizing the resources to\nexplore the relevant parts of the phase space. We showed that DDMD\nestimates the free energy landscape of the RNA stem-loop reasonably\nwell at room temperature. Our simulation framework runs at a constant\ntemperature without external biasing potential, hence preserving the\ninformation on transition rates, with a computational cost much lower\nthan that of the simulations performed with external biasing potentials.\nWe also introduced a reweighting strategy for obtaining unbiased free\nenergy surfaces and presented a qualitative analysis of the latent\nspace. This analysis showed that the latent space captures the relevant\nslow degrees of freedom for the RNA folding problem of interest. Finally,\nthroughout the manuscript, we outlined how different parameters are\nselected and optimized to adapt DDMD for this system. We believe this\ncompendium of decision-making processes will help new users adapt\nthis technique for the rare-event sampling problems of their interest.