Fig. 5.

Evaluation of helical tracing performance by determination of precision and recall. (A) Helical traces superimposed on micrograph obtained by interactive helix tracing. Great care was used in the procedure such that the results were considered the ground truth. (B) Helical traces determined by automated tracing procedures MicHelixTrace or RELION. (C) Grey contour areas present an inflated and cropped area of interactively picked helix traces considered ground truth. Automated tracing results were mapped onto grey contours to determine the precision P. The overall number of pixels that do or do not coincide with the ground truth can be evaluated to determine true positive and false positive traces and thus the precision P. High precision P indicates that a large fraction of the found helices correspond to true helices. (D) Grey contour areas here present inflated and cropped automated tracing results. Interactively picked helices (ground truth) were mapped onto grey contours. Pixels that overlap the contour areas were correctly found as true positives, otherwise they are false negatives. High recall R means that a high proportion of the ground-truth helices were found. (E) Comparison of averages from in-plane rotated power spectra of TMV obtained from interactively (left) and automatically (right) traced segments by MicHelixTrace. Inset: Magnified area between 0.05 and 0.15 1/Å in y-direction and between 0.1 and 0.2 1/Å in x-direction shows that layer lines are more discrete and less blurred when computed from automatically traced data set due to more precise in-plane angle alignment. (F) Normalized layer line intensity plot 1/7.34 Å confirms that layer line from automated tracing decays slower towards higher resolution compared to interactive tracing.