journal menu
![[Figure 4]](id5015fig4mag.jpg)
|
|
Figure 4
Threshold dependence and connectivity filtering. (a) Illustration of off-target segments in membrane-distal regions of nanodisc-embedded D. melanogaster NOMPC (EMDB entry EMD-8702). Density contours for different binarization thresholds are shown. (b) 2D section views at the three section planes indicated in (a) show the overlap of the unfiltered micelle prediction with the target map. (c) Effect of filtering based on largest connected volume on the micelle-subtracted volume. The top row shows how subtraction of the predicted segment without filtering also removes density from the off-target segments. The bottom row shows how segment filtering restricts subtraction to only the micelle density. (d) F1 score as a function of binarization threshold evaluated on the held-out test dataset. Segmentation performance peaks at an intermediate threshold, reflecting the trade-off between recall at low thresholds and precision at high thresholds. The position and sharpness of this optimum vary across reconstructions. The curve for EMDB entry EMD-8702 is shown in red. (e) Schematic illustration of the interactive thresholding workflow implemented in SURFER, highlighting how users can explore the effect of threshold choice on segmentation quality and apply connectivity filtering prior to density subtraction. |
![[Figure 4]](id5015fig4.jpg)
 | STRUCTURAL BIOLOGY |
ISSN: 2059-7983
access
