Supplementary Figure 6: Comparison of the resolution and estimated tilting angle by different programs for the 40^o-tilted dataset of the influenza HA trimer²⁷.

a) FSC curves corresponding to different method of local defocus determination. Two tests were performed to evaluate the defocus refinement by THUNDER. For the first one, CTFTilt¹⁸ was used to obtain the initial defocus estimation, and then THUNDER with defocus refinement was used to reprocess the dataset. The final resolution of the reconstruction was reported at 3.69 Å with most side-chains visible (Fig. 5b). The defocus refinement contributed a 0.9 Å resolution improvement (see the red dotted and solid lines). Another test was performed with the defocus value (estimated by GCTF²⁶) from the downloaded parameter file (EMPIAR entry code: 10097), and reported 3.81 Å resolution before the defocus refinement. This result is better than the published one, but slightly worse than the one above, indicating less defocus accuracy of GCTF than that of the defocus refinement by THUNDER. Unexpectedly, the defocus refinement on the basis of the GCTF results did not obviously improve the resolution. Considering the defocus refinement is just a local refinement, a possible reason is that the initial defocus error of some particles is beyond the working range of the defocus refinement. b) Histogram of tilt angles of micrographs. To further validate the estimated defocus, we made statistics of the tilt angles of micrographs, measured by plane fitting of the per-particle defocus values from the same micrograph. The histograms corresponding to CTFTilt and the defocus refinement of THUNDER are pretty identical and closed to the expected 40^o tilt. The one from GCTF (from the downloaded parameter file) has a different angle distribution from the other three, which might imply errors in measured defocus.