Fig. 2: RNA structures from AFM images.

a–c, Raw (a) and processed (b) AFM particle images and crystal structure (PDB: 2A64) rendered as a molecular surface (c) of the full-length RPR. The raw single-molecule images were taken once—that is, without signal averaging. For particles P1, P2 and P3, the crystal structure were orientated to best match the AFM images. Scale bar, 10 nm. d, Resolution estimation using the auto-correlation value³⁴ (ACV). Inset graphs show similar plots for the spatial regimes with an abrupt decrease in ACV, where the discontinuities in the ACV profiles indicate structural features present in the image at a particular spatial resolution, observed at approximately 0.87 nm⁻¹ (11.5 Å), 0.90 nm⁻¹ (11.1 Å) and 0.80 nm⁻¹ (12.5 Å) in P1, P2 and P3, respectively. e, The first derivative of ACV showing more detailed profile variation. Grey bars indicate ACV discontinuity regimes, with a maximum spatial threshold at approximately 0.34 nm⁻¹ (29 Å) and minimum spatial thresholds of approximately 0.87 nm⁻¹ (11.5 Å), 1.4 nm⁻¹ (7.2 Å) and 0.97 nm⁻¹ (10.3 Å) for P1, P2 and P3, respectively. f, AFM topographical images of the RPR catalytic domain (PDB: 3DHS) simulated at different noise levels. The z height signal profiles in the x–y plane are shown below each image. Applied noise levels (left to right): 5, 10, 15, 20, 30, 40 and 50% of the maximum z height. g, DCCMs of free (left) and AFM-restrained (right) molecular dynamics trajectories. h, Stacked r.m.s.d. relative to the ground-truth structures (r.m.s.d._GT) distribution plots of the dynamic fitting trajectory in Å at each noise level.