The Multi-Target Model

A multi-color experiment images two or more targets — say an IgG variant and complement C1q — each in its own channel. SMLMAnalysis handles these by generalizing the single-channel pipeline, not by bolting on a separate system. If you understand The Pipeline Model, you already understand most of this.

Two axes of generalization

The single-channel pipeline threads one BasicSMLD through a list of step configs, dispatching analyze(state, config) on their types. The multi-target pipeline keeps exactly that idea and changes only two things:

Everything else — the analyze(state, step) dispatch, the (result, info) tuple per step, the threaded state — is unchanged. A cross-channel step is simply an analyze(::Vector{BasicSMLD}, ::AbstractMultiTargetStep) method.

Two phases

One MultiTargetConfig and one analyze call describe a run with two phases:

1. Per-channel pipelines. Each channel is analyzed independently by a full single-channel pipeline — the orchestrator literally calls analyze(data, AnalysisConfig) once per channel, reusing that channel's own camera, steps, ROI, and verbosity. The resulting per-channel SMLDs become the channel vector.
2. Cross-channel steps. The ordered AbstractMultiTargetSteps then run over that Vector{BasicSMLD}, threading it from one step to the next.

Phase 1 shares no state between channels — each is exactly the analysis you would run on a single color. All channel interaction happens in phase 2.

Pass-through vs. state-modifying steps

As with single-channel steps, a cross-channel step either transforms the state or just reads it for its outputs:

• Pass-through steps return the channel vector unchanged — Composite Render (the multi-color overlay image) and Cross-Correlation (the co-localization g(r)).
• State-modifying steps replace the vector — Cross-Alignment returns aligned SMLDs. Because the state threads in order, a composite render placed after an alignment step shows the corrected overlay, while one placed before shows the raw registration. Step order is meaningful.

No provenance is lost

analyze returns a (MultiTargetResult, MultiTargetInfo) tuple that exposes both the merged channel vector and each channel's complete single-channel records:

result.smlds         # Vector{BasicSMLD} (aligned if Cross-Alignment ran)
result[:IgG]         # the channel's full AnalysisResult
info.channels[:IgG]  # the channel's full AnalysisInfo

Because every channel keeps its own AnalysisResult and AnalysisInfo, you can reach any per-channel step's info exactly as in a single-channel run — see Data Model & Provenance.

For the concrete configuration, the analyze call shape, the on-disk output layout, and each cross-channel step in detail, see Multi-Channel Analysis.