Multi-Channel Analysis

A multi-color experiment images two or more targets in separate channels — for example an IgG variant and complement C1q, each on its own dye. This page shows how to analyze every channel with its own pipeline and then run cross-channel steps (composite renders, alignment, cross-correlation) that operate on all channels together. The orchestrator is MultiTargetConfig; if you are new to the single-channel flow, read The Pipeline Model and Running a Pipeline first.

The MultiTargetConfig

MultiTargetConfig is the single object that describes a multi-channel run — the channels, their colors, and the cross-channel steps to run once each channel is analyzed. It is the orchestrator you construct and hand to analyze:

MultiTargetConfig <: AbstractSMLMConfig

mt = MultiTargetConfig(
    labels = [:IgG, :C1q],
    steps = [
        CompositeRenderConfig(zoom=20.0, strategy=GaussianRender()),
        CrossAlignConfig(method=:entropy),
        CompositeRenderConfig(zoom=20.0, strategy=GaussianRender()),
    ],
    outdir = "output/cell1/",
)

Its fields in practical terms:

mt = MultiTargetConfig(
    labels = [:IgG, :C1q],
    colors = [:cyan, :magenta],
    steps = [
        CompositeRenderConfig(zoom = 20.0, strategy = GaussianRender()),
        CrossAlignConfig(method = :entropy),
        CompositeRenderConfig(zoom = 20.0, strategy = GaussianRender()),  # post-alignment
        CrossCorrConfig(r_max = 0.5, dr = 0.005),
    ],
    outdir = "output/cell1/",
)

The steps run in order, threading a Vector{BasicSMLD} through. Most steps pass the SMLDs through unchanged; Cross-Alignment replaces them with aligned copies, so a second composite render placed after it shows the corrected overlay.

The shape of a multi-channel run

A multi-channel analysis has two phases, both described by the one MultiTargetConfig above and run by a single analyze call:

1. Per-channel pipelines. Each channel is analyzed independently with its own AnalysisConfig — exactly the pipeline you would run on a single color.
2. Cross-channel steps. Once every channel has produced a BasicSMLD, the ordered multi-target steps run across the resulting Vector{BasicSMLD}.

Running it

Call analyze with a vector of (data, AnalysisConfig) tuples — one per channel, in the same order as labels — together with the MultiTargetConfig:

(result, info) = analyze([
    (image_stacks_647, config_647),
    (image_stacks_568, config_568),
], mt)

Each data is whatever the single-channel analyze accepts (an image stack, a vector of stacks, or a file path). Each channel's AnalysisConfig carries its own camera, steps, ROI, and verbosity; the orchestrator reuses all of those but redirects the channel's output under outdir/<label>/, so you do not set a per-channel outdir yourself.

Working with the result

analyze returns a (MultiTargetResult, MultiTargetInfo) tuple, following the JuliaSMLM convention (see Data Model & Provenance).

MultiTargetResult fields and indexing:

result.smlds        # Vector{BasicSMLD}, one per channel (aligned if CrossAlign ran)
result.channels     # Dict{Symbol, AnalysisResult}
result.step_infos   # Vector{StepInfo} for the cross-channel steps
result.outdir       # root output directory
result[:IgG]        # AnalysisResult for one channel (indexing == result.channels[:IgG])
result[:IgG].smld   # that channel's final SMLD
keys(result)        # the channel labels, in order

MultiTargetInfo carries the metadata:

info.elapsed_s      # total wall time, seconds
info.channels[:IgG] # AnalysisInfo for one channel's pipeline
info.step_infos     # Vector{StepInfo} for the cross-channel steps
info.steps          # Dict{Symbol, Any}: step name => upstream info

Cross-channel steps

Each multi-target step is a <: AbstractMultiTargetStep config dispatched on the channel Vector{BasicSMLD}. They are documented on their own pages — add them to MultiTargetConfig.steps in the order you want them to run:

• Composite Render — overlay the channels into a single multi-color image, each tinted by its colors entry (per-step colors override the MultiTargetConfig default). Accepts the same strategies as Rendering (GaussianRender, HistogramRender, CircleRender). Pass-through.

• Cross-Alignment — remove residual channel-to-channel shift (chromatic offset, registration error) by entropy or FFT cross-correlation. State-modifying: it returns aligned SMLDs, so steps after it see the corrected positions.

• Cross-Correlation — quantify co-localization between a pair of channels via the pair cross-correlation g(r). Pass-through.

Output layout

The run mirrors the two-phase structure on disk: a subdirectory per channel, a composite/ directory for the cross-channel steps, and per-channel SMLD files plus a serialized config at the root.

output/cell1/
├── IgG/                    # per-channel pipeline output (01_detectfit/, 02_filter/, ...)
├── C1q/
├── composite/              # cross-channel step outputs, numbered in step order
│   ├── 01_compositerender/
│   ├── 02_crossalign/
│   ├── 03_compositerender/
│   ├── 04_crosscorr/
│   └── README.md           # color scheme, per-channel counts, step summary
├── smld_IgG.h5             # per-channel saved SMLDs (with drift model)
├── smld_C1q.h5
└── multi_target_config.toml