Batch and specialised analysis

Requires a license.

Batch analysis

The point of batch analysis is that a gating strategy is a decision you should make once and apply consistently — not re-draw per sample, where it drifts.

The gate tree is shared by every open file, so this happens naturally: a gate you draw on one file applies to all of them.

1. Open every file in the experiment.
2. Build and check the strategy on one representative sample.
3. Click Batch in the toolbar (shown once more than one file is open and at least one gate exists), or Open Batch Analysis from the command palette.
4. Read the results as one table: a row per open file, a column per gate, for the statistic you pick under Metric: — Count, % of Total, % of Parent, or a per-parameter Mean, Median, Geo Mean, Std Dev or CV%. The foot of the table gives the mean and SD across files.

Export Matrix saves that table as CSV; Export Full saves every statistic for every file, gate and parameter. Each file is computed with its own compensation.

To evaluate the gates on only some files, Ctrl+Click them in the file list and choose Apply Gates.

Check a few samples individually afterwards. A strategy that fits the sample you built it on can fail on a sample with different staining intensity or a shifted population, and a batch table looks equally confident either way. Spot-checking the outliers is the whole discipline here.

Dose-response

For experiments where samples are a concentration series: extract a population frequency or channel statistic per dose and fit the response.

From gated data: put the samples in one group, pick the gate and the readout — % of Parent, % of Total, or a median, mean or geometric mean of a channel inside the gate — and enter each sample’s concentration. The readouts are computed the same way as the statistics table, then fitted with a four-parameter logistic (4PL) curve. You can also type values in directly.

The fit reports Bottom below Top; the sign of the Hill slope gives the direction (positive rises, negative falls). If you constrain the slope to a direction the data do not follow, the panel warns you rather than reporting a curve with its ends swapped.

Proliferation analysis

For dye-dilution proliferation experiments — CFSE and similar — where successive generations appear as halving peaks in a single channel.

Choose the dye channel, the population (gate it to live single lymphocytes first: debris and doublets otherwise look like extra generations) and Max Generations, the most generations to look for. Each division halves the dye, so the app finds the undivided peak and fits the generations at the evenly spaced positions that halving predicts, refining the spacing from your data. Events dimmer than the last generation looked for are reported as unassigned.

It reports events per generation and the standard indices (Roederer; as in FlowJo), computed from precursor frequencies — the events in generation i divided by 2ⁱ:

Index	Meaning
% Divided	Share of the original cells that divided at least once
Division Index	Average number of divisions of all original cells
Proliferation Index	Average number of divisions of the cells that divided
Expansion Index	Fold expansion of the culture
Replication Index	Fold expansion of the cells that divided

Versions before 0.2.2 weighted generations by event count, which gives different values; compare results from 0.2.2 onward with each other, not with earlier ones.

Graph builder

For plots the standard views do not cover — arranging populations and statistics into the figure you actually want rather than the one the layout offers.

Its statistical tests give the same results as R: Welch’s t-test; Mann-Whitney, exact when both groups have fewer than 50 values, as in R 4.6, otherwise the normal approximation with continuity and tie correction; Kruskal-Wallis with tie correction; and one-way ANOVA, reported as F with both degrees of freedom. The panel says which p-value method was used.

With three or more groups, choosing the t-test runs a one-way ANOVA, and choosing Mann-Whitney runs Kruskal-Wallis. Each is followed by pairwise comparisons: Welch’s t-tests after the ANOVA, Mann-Whitney tests after Kruskal-Wallis, with Bonferroni-adjusted p-values. These are valid and conservative, but they are not GraphPad Prism’s defaults (Tukey after ANOVA, Dunn’s after Kruskal-Wallis), so pairwise p-values will differ from Prism’s. State the method in your write-up.

Practical sequence for an experiment

1. Open every file in the experiment.
2. Fix channel names once — see the parameter editor in The interface.
3. Compensate, from controls where you have them.
4. Build the strategy on one representative sample, and check it.
5. Batch it across the rest.
6. Spot-check the outliers by hand.
7. Export statistics, the compensation matrix, and the workspace.