I am processing a helical filament formed by bacterial tubulin-like proteins. The expected structure consists of four distinct protofilaments arranged helically along the filament axis.
I previously processed this dataset using CryoSPARC v4.0.1. The reconstruction obtained in v4.0.1 showed the expected architecture, with four clearly separated protofilaments following a helical path.
After upgrading to CryoSPARC v5.0.6, I reprocessed the same dataset several times. However, the reconstruction consistently converged to a tube-like or cylindrical density. Instead of four separated protofilaments, the densities of the four protofilaments appear to merge laterally into a continuous cylinder.
The same raw dataset was used in both versions, and I attempted to keep the processing workflow and major parameters as similar as possible. The main difference is the CryoSPARC version.
I would like to ask:
Were there any changes to the helical reconstruction or refinement algorithms between v4.0.1 and v5.0.6 that could affect this type of reconstruction?
Are there any new default parameters or settings in v5.0.6 that should be carefully checked when processing an unusual helical assembly with only four protofilaments?
Could the cylindrical reconstruction result from changes in helical symmetry optimization, regularization, initial model handling, or alignment settings?
Has anyone observed a similar difference between CryoSPARC v4 and v5 when processing helical filaments?
The approximate helical parameters of this specimen are:
Helical rise: approximately 79–82 Å
Helical twist: approximately −5.5°
Number of protofilaments: four
Protein type: bacterial tubulin-like protein
Expected morphology: four separated protofilaments forming a helical tube
There’s a large version gap between these two: v4.0.1 was from late 2022, while v5.0.6 is from early 2026, so a good amount has changed across many stages of the processing pipeline. Helical reconstruction is also quite sensitive to a range of details spread across the whole workflow (initial models, starting symmetry estimates, particle curation, masking, etc) some of which won’t be held constant between two separate re-processings of the same data, even with fixed random seeds. Given this, I’d want to isolate the variables as much as possible before attributing the cause to different CS versions.
To help narrow this down, a few questions:
In v5.0.6, did you reprocess from scratch, or start from the extracted particles already available from the v4.0.1 run? I.e. were the particles the same set, or a newly curated/re-extracted one?
Roughly what resolutions did you obtain in each version? Did either run reach better than a claimed ~4 Å? If so, did either density have interpretable features incl. secondary structure?
Do you still have access to the original v4.0.1 project and its results?
If you do still have the v4.0.1 project, the most informative test would be to remove as many confounds as possible and compare like-for-like:
Take the inputs to your final v4.0.1 helical refinement (the same volume, particles, and mask).
Connect those identical inputs to a fresh Helical Refinement job in v5.0.6.
Run with parameters matched as closely as possible to the v4.0.1 job, including a fixed random seed.
Clone and run it a few times (3 is usually enough) to gauge residual run-to-run variance.
If v5 reproduces the cylinder from the exact same inputs that gave separated protofilaments in v4, that points at an algorithmic or default-parameter change worth digging into. If it instead recovers the protofilaments, the difference more likely lives upstream in the two workflows (e.g. discrepancies in particle curation, particle picking, initial models, etc.)
Happy to help interpret the outputs once you’ve had a chance to run this.