view article
Figure 2
Quantitative comparisons between Smart Leginon Autoscreen (SLA) and expert microscope operators (Op). Autoscreen and operators each independently selected what they considered to be the `best' 4 squares, where for Autoscreen each square was selected from 4 equally spaced square area ranges spanning all square areas (one grid atlas is shown on the left; see Fig. S8 for all atlases). Autoscreen took <10 min for the operator to set up and then run in a completely unattended manner for 5.4 h. The operators spend an average of 6.0 h to screen 11 grids (*extrapolated from 3 grids), of which most of the time is spent operating the microscope, interspersed with several short periods (5–10 min per grid) of time away from the microscope. (Note: calculations assume that the operators do not take any breaks away from the microscope.) In terms of percentage of good holes available from hole magnification images, Autoscreen (95.9% good holes) performed better than the average from the operators (90.6% good holes) and comparable to the best operator (95.2% good holes). Figs. S9–S14 show visual analyses of all holes and Table 1[link] shows the quantifications. From the random holes targeted (SLA) and central holes (Op), CTF resolution estimation for Autoscreen holes (7.3 ± 2.6 Å) was comparable to the average obtained by operators (7.4 ± 2.9 Å). Estimates of ice thickness show comparable values between Autoscreen (32.9 ± 7.1 nm) and operators (35.0 ± 18.7 nm). Table S1[link] shows the raw data. The hole magnification image most representative of the average in terms of hole quality is shown in the last row. Targets are shown by white squares and bad holes are shown by red circles with lines through them. Note: the representative image for operator 4 is a composite image due to there being no image that closely represents the average.

IUCrJ
Volume 10| Part 1| January 2023| Pages 77-89
ISSN: 2052-2525
https://doi.org/10.1107/S2052252522010624