I am trying to use the Filament Tracer (cryoSPARC version 5.0 beta) on a difficult dataset. Vast majority of the protomers disassemble and only few filament-like feature can be manually picked. I was manually picking on denoised micrographs, and tried the Filament Tracer on both regular and denoised micrographs. Filament tracer definitely overlooks a nice filament on the left micrograph and picks some random two positions on the right one instead. How would you tune the parameters to increase the chance of hitting the rare but relatively visible filaments?
usually you can pick more generously by lowering the hysteresis thresholds like below
For difficult cases we actually use template matching (or blob picking, your templates don’t look very defined) instead of filament tracing. Don’t overdo the overlap to prevent inflated FSCs.
For really precise filament particle picking we use CrYOLO and import the coordinate starfile. I attached DNA-bound Par filaments below, they are really sparse. Training from 20 manually picked micrographs resulted in great picking accuracy.
Your example micrographs look very dark, are they lowpass filtered? If not, you might need to fine-tune ice quality (gridsquares with thinner ice), lower particle concentration, reduce ethane contamination etc.
