The JuliaSMLM Ecosystem

SMLMAnalysis sits at the top of JuliaSMLM, a set of focused packages that each own one part of the SMLM analysis problem. This page is the map: what each package does, which step it backs, and when you might call it directly instead of through the pipeline.

All packages share the core data types from SMLMData (BasicSMLD, Emitter2DFit, cameras), so values pass between them without conversion.

Dependency map

SMLMData ............ core types (Emitter, Camera, BasicSMLD, ROIBatch) — no deps
   │
   ├── SMLMBoxer ............ ROI detection from raw frames
   ├── GaussMLE ............. GPU-accelerated MLE PSF fitting
   ├── SMLMFrameConnection .. link blinks across frames (+ uncertainty calibration)
   ├── SMLMDriftCorrection .. fiducial-free drift correction
   ├── SMLMRender ........... super-resolution image rendering
   ├── SMLMBaGoL ............ Bayesian grouping of localizations (RJMCMC)
   ├── SMLMClustering ....... DBSCAN / Hierarchical / Voronoi / Hopkins
   ├── SMLMSim .............. simulation + image generation
   └── MicroscopePSFs ....... PSF models (Gaussian, Airy, …)
                              │
                       SMLMAnalysis (integrates all of the above)

What each package does

Package	Backs the step	Role	Primary reference
SMLMBoxer	Detection	Finds candidate molecules and cuts ROIs from raw frames	Huang et al. 2013
GaussMLE	Fitting	Maximum-likelihood Gaussian/astigmatic PSF fitting with CRLB uncertainties (GPU)	Smith et al. 2010
SMLMFrameConnection	Frame Connection	Links repeated blinks of one fluorophore into single, higher-precision localizations; optional uncertainty calibration	Schodt & Lidke 2021
SMLMDriftCorrection	Drift Correction	Fiducial-free drift estimation by entropy minimization	Cnossen et al. 2021; Wester et al. 2021
SMLMRender	Rendering	Turns localizations into super-resolution images (histogram / Gaussian / circle / ellipse)	—
SMLMBaGoL	Bayesian Grouping	Groups localizations into true emitters via RJMCMC, beyond raw precision	Fazel et al. 2022
SMLMClustering	Clustering	Cluster labeling (DBSCAN/Hierarchical/Voronoi) and spatial statistics (Hopkins, Voronoi density)	Ester et al. 1996; Levet et al. 2015
SMLMSim	—	Simulates SMLM data and generates synthetic image stacks (used throughout the docs)	—
MicroscopePSFs	—	PSF models used by fitting and simulation	—

Full citations are on the References page; each step page repeats its own primary reference and links to the upstream package's documentation for the algorithm details.

Steps SMLMAnalysis owns

Not every step comes from an upstream package. SMLMAnalysis implements several of its own — the parts of a real pipeline that do not belong to any single lower-level package:

Quality Filter (FilterConfig) — threshold filtering on photons, precision, p-value, PSF width, and z.
Intensity Filter (IntensityFilterConfig) — Poisson upper-tail rejection of multi-emitter events against an estimated excitation field.
Density Filter (DensityFilterConfig) — removal of isolated localizations by neighbor count, with automatic threshold selection.
Multi-channel steps — Composite Render, Cross-Alignment, and Cross-Correlation operate across colors and are built here on top of SMLMRender / SMLMDriftCorrection / NearestNeighbors.

When to use a package directly

The pipeline is the right tool when you want a reproducible, multi-step analysis with provenance and on-disk diagnostics. Reach past it, straight to an upstream package, when:

You need a capability the pipeline doesn't expose — an upstream function or option SMLMAnalysis doesn't surface. Because all packages share SMLMData types, you can pull result.smld out of a pipeline and pass it to any upstream function.
You are developing or debugging one algorithm in isolation and want its package's full API and docs.
You are doing post-hoc exploration on an already-analyzed SMLD (e.g. trying several clustering parameters) and don't need it recorded as a pipeline step.

Since using SMLMAnalysis re-exports the key upstream types and verbs (cluster, render, run_bagol, frameconnect, the config types, the emitter types), you can usually do both styles from the same session without extra imports.