Extended Data Fig. 7: Structure calculation using FARFAR2 prediction S257 as starting model for RNase P RNA catalytic domain.
From: Determining structures of RNA conformers using AFM and deep neural networks
a, HORNET-UML analysis of non-convergence for S257. Even though S257 shows a good FARFAR2 score and the best ARES score, this structural model exhibits a fold completely different from the GT structure. We applied HORNET to S257 using simulated AFM topography of the GT structure (PDB 3DHS with missing residues modeled) as the driving force for dynamic fitting. For the trajectory time of ~11μs (estimation using CafeMol31 unit time of 49 fs), the structure could not converge on a conformation close to the target structure, showing a minimum true RMSD of 21.4 Å and a major population centered at 27.5 Å RMSD in the final cohort of models (gray). b, Actual RMSD vs estimated RMSD using HORNET for the S257 dynamic fitting trajectory, the minimum estimated RMSD using HORNET, for the entire trajectory of ~16 million models, is 12.4 Å. However, this minimum is a clear outlier in the distribution, which is centered at ~27 Å. HORNET was able to show that the trajectory of S257 could not converge to a structure resembling the GT structure. c, radio of gyration (Rg) as function of a coarse-grained MD simulation of ~2.9 \(\mu \)s trajectory using CafeMol31. This simulation was performed to optimize the energy minimization of S257. d, To select the minimized structure from the MD trajectory, we applied the HORNET-UML approach without the AFM restraint, and the cluster with structures having the lowest Total and Go energies were selected (top). Next, we selected the five models having the lowest total energy (star symbol) with RMSDs relative to the GT structure of 19.3, 17.8, 19.0, 19.2 and 14.4 Å (below). e, Correlation between the true and estimated RMSDs by HORNET relative to the GT structure for all models from the five independent trajectories using the five selected the lowest-total-energy. The black dots indicate the top 10 HORNET-derived models having the lowest estimated RMSD. f, RMSD of the top 10 HORNET-derived models vs the GT structures.
