High-dimensional analysis

Define the input

Choose the samples, parent population, and markers for the analysis. Review compensation or unmixing, transforms, and sample quality first. Record these choices so that you can distinguish preprocessing changes from changes to the analysis method.

Confirm that the selected markers are available across your samples. Consider whether acquisition channels, time, or other technical parameters belong in the selected feature set for your question.

Run an embedding

1. Open Spectral & high-dim and select the dataset or cohort to inspect.
2. Choose an embedding method, such as PCA, UMAP, or t-SNE, and review its parameters and seed.
3. Run the analysis. Inspect the resulting view colored by sample, population, or marker expression.
4. Check the reported event and drawing counts. Keep track of the input used for the computation and what is displayed in the plot.

Review clusters

Use the clustering panel for methods such as FlowSOM or PhenoGraph. Review the selected method and its settings before running. Inspect marker expression and sample composition for each cluster, then compare it with the relevant gated populations.

Treat cluster labels as analysis outputs that need interpretation. Verify a proposed cell identity against the marker measurements and experimental context before reporting it.

Record the analysis choices

Keep the marker list, input population, sample selection, preprocessing, method parameters, and seed with the results. When you rerun an analysis, change one choice at a time where possible and compare the outputs before deciding what to report.

Keep computation and display scope separate

Record how many events were eligible, how many entered the computation, and how many were drawn when those counts are shown. A plotting limit and an analysis subsample are different choices. Keep sample identifiers available when reviewing a combined dataset.

Compare sample representation in the output with the selected input. If one sample contributes most of the events, make that visible in the analysis record rather than describing the result as a balanced cohort by default.

Review a rerun systematically

1. Preserve the original sample selection, parent population, markers, preprocessing, method parameters, and seed.
2. Change one setting and record its previous and new values.
3. Compare marker expression and sample membership for the populations of interest.
4. Keep cluster labels associated with their own run; a label such as cluster 3 does not guarantee the same membership after a rerun.
5. Export the final figures with the settings needed to identify the computation.