Flow Cytometry Data Analysis

Intended use case

Use this page as a practical reading order when your dataset or scientific question sits near exploratory flow-cytometry-style analysis. The current docs support this topic mostly through multivariate figures, clustering-style views, and compositional caution pages.

This is not a claim that the product exposes a ready-to-run flow-cytometry workflow. It is a cautious bundle of nearby pages that can help you review structure, heterogeneity, and warning signals.

In other words, this page is a map to the outputs that are currently closest to exploratory flow-cytometry work: reduced-dimension figures, cluster-style overlays, matrix views, and caution checks about compositional structure or batch-like design problems. It should not be read as a promise that Licklider will automatically identify cell populations, validate gating strategy, or produce one canonical flow-cytometry report.

Suggested steps

1. Start with the figure pages that help you inspect broad structure across many variables.
2. Use PCA-style and clustering-style views as exploratory screens, not as proof that the data contain one true set of populations.
3. Review matrix-style displays next when feature-by-sample patterning matters more than one reduced two-axis projection.
4. Check compositional and batch-related cautions before treating the visual pattern as a stable biological story.
5. Review limitations before promoting an exploratory pattern into a stronger public claim.

This ordering is deliberate. The current support surface is stronger for exploratory pattern-reading and warning disclosure than for one end-to-end domain-specific pipeline, so the page leads with structure screens and guard pages rather than promising an automated conclusion.

Suggested methods

• Compositional Data Analysis is the closest current method-adjacent page when the data behave like parts of a whole rather than ordinary unconstrained measurements.

This workflow page does not itself define a flow-cytometry-specific statistical method. Its role is to help you decide which existing figure or caution page is the safest next read for your current dataset.

Suggested figures

• PCA Biplot gives a two-component score-style view for asking whether samples look broadly close, separated, or unusual in one reduced projection.
• K-means Clustering Plot gives a cluster-colored two-axis overlay for inspecting one chosen partition, not for proving that one true set of populations has been discovered.
• Heatmap gives a color-coded matrix view when the main output you need is pattern comparison across many features and samples rather than one reduced scatter display.
• Hierarchical Clustering Heatmap should be read narrowly as a dendrogram-first grouping view because current support is weaker than a full integrated clustered-heatmap workflow.

These are the concrete outputs this page can currently guide you toward. They are exploratory figure surfaces, not a single domain-specific verdict such as "the correct cell populations were found."

Required checks

• Compositional Data Warning matters when fractions, abundance-like values, or near-constant totals could change the meaning of the analysis. When multiple signals align, Licklider can automatically flag potentially compositional structure and ask you to confirm whether the result should carry a disclosure or be treated as descriptive only.
• Batch and Plate Confounding Detection is worth reviewing when batch, plate, run-order, or similar processing columns are present. Licklider can check the encoded table structure for overlap between batch-like variables and experimental groups, but it cannot detect hidden processing structure that is missing from the uploaded table.

Those checks are important to the "how" of safe use. The current product is more reliable when it surfaces a caution early than when a reader assumes the visual pattern is safe by default.

Final artifacts

• A cautious outcome today is usually a bundle of exploratory outputs: for example, a PCA-style projection, a cluster-colored overlay, a heatmap or dendrogram-style grouping view, and any relevant warning or disclosure about compositional structure or batch-like confounding.
• Readers should expect caveat-aware interpretation and next-step review rather than one automated flow-cytometry conclusion.
• This page is therefore useful when you want to know what Licklider can currently show, warn about, and help you review around a flow-cytometry-adjacent question.

Notes on interpretation

• PCA and clustering views here are exploratory.
• They can help organize what to inspect next, but they do not automatically discover stable cell populations.
• Heatmap and hierarchical displays are also interpretation-heavy because scaling, ordering, and feature choice strongly shape what the viewer sees.

If the apparent structure changes substantially after changing scaling, feature inclusion, clustering choices, or preprocessing, that instability is part of the interpretation rather than something this page resolves for you automatically.

What Licklider does not detect for you

• Licklider does not automatically determine the scientifically correct number of clusters or the one true set of biological populations.
• Licklider does not automatically determine whether the currently displayed separation is robust to preprocessing, scaling, feature choice, or projection choice.
• Licklider does not automatically infer hidden batch, plate, operator, acquisition, or handling structure when that metadata is absent from the uploaded table.
• Licklider does not automatically tell you that a visually strong exploratory pattern is already suitable for claim-bearing biological interpretation.

These boundaries matter because flow-cytometry-style analyses are easy to over-read. A cluster-like pattern can reflect preprocessing choices, composition constraints, or acquisition structure rather than one stable biological story.

Short cautions

• If batch, plate, or acquisition effects are plausible, review them before treating visible clusters as biology first.
• If values behave compositionally, keep Compositional Data Warning nearby.
• For current support boundaries, also see Known Limitations.

What this page does not cover

• Flow-cytometry-specific gating strategy, compensation, raw event processing, or population calling.
• A guarantee that the linked clustering-style figures discover stable cell populations automatically.
• A replacement for assay-specific preprocessing, QC, or domain review before making a stronger scientific claim.
• Full details of each linked figure or warning page; use the links above when you need the concrete contract for one output surface.

TODO (Phase02+)

• Expand only if the public product surface later confirms a stronger flow-cytometry-specific workflow contract.