Cell-Cycle Modeling

6. Cell Cycle Modeling

Once a channel is plotted, click Cell Cycle Modeling in the sidebar. The file and channel controls slide away and the sidebar reveals the Pre-modeling QC Filtering box (four filters: Structural, Time, Cell Gate, Singlet Gate — toggle individually or with Run All), the Identify Peaks panel, and the Model & Fit panel; moving these out of the plot panel also gives the histogram more height. QC is a separate action, applied on its own before peak review. Back returns the sidebar to the file/channel controls without losing any progress.

With exactly one sample checked, Identify Peaks reviews that sample; with several checked at once, click a row in the table to choose which one it's reviewing. Detect Peaks runs automatic peak detection and proposes G1 and G2/M regions (when several samples are checked and none is singled out, it detects for all of them); the four numeric fields let you adjust each region's left/right limits directly, with inline validation if an edit breaks their ordering. Reset discards manual edits and restores the automatic proposal; Accept marks the current regions reviewed; Apply to All copies the reviewed sample's regions to every plotted sample.

In Model & Fit, pick a model (Dean–Jett, Dean–Jett–Fox, Watson Pragmatic, or Automatic, which chooses between Dean–Jett and Dean–Jett–Fox) — the dropdown opens on Model Selection… so the choice is always deliberate. Fit Current fits the reviewed sample; Fit All Samples auto-detects, averages the regions across every plotted sample, and fits each. Fitted G1/S/G2–M fractions are written to the metadata table (one column group per model in use), alongside the accepted G1/G2 peak-region bounds.

Watson Pragmatic is a pragmatic decomposition, not a likelihood fit: it estimates G1 and G2/M locally from each peak's clean (uncontaminated) flank and then takes S phase as whatever observed signal is left over. That residual S is measured strictly between the two fitted peak centers — signal below the G1 center (sub-G1 debris) or above the G2/M center (post-G2 aggregates) is never counted as S. Watson models no separate debris/aggregate component, so keep such contaminants outside the G1–G2/M interval. Because it is a decomposition rather than a generative model, its results are deliberately never ranked against Dean–Jett or Dean–Jett–Fox by AIC/BIC.

Pre-modeling QC filters

Automated preprocessing works best in layers rather than a single step: first remove acquisition artifacts over time, then apply cell-cloud (debris) gating and explicit singlet/doublet discrimination, and only then fit the histogram. Order matters because clogs, flow-rate surges, dynamic-range artifacts, and margin events distort the very populations the model is trying to quantify. PhaseFinder's four QC filters implement that order:

1. Structural QC rejects non-finite, negative, and saturated measurements while preserving original event indexes.
2. Time QC removes unstable acquisition periods. When you switch it on, you choose a method: Robust summary QC monitors each acquisition bin's event rate, medians, and IQRs; Peak-tracking QC follows the major distribution peaks across overlapping acquisition bins and removes regions where those peaks shift abnormally. The panel then reports the method, events removed, and the acquisition regions it flagged. Use Change… to switch methods and re-run.
3. Cell Gate fits an FSC/SSC cell-cloud gate. Drag the teal ellipse or its center marker to reposition it, hold Shift or Ctrl while dragging (or Ctrl plus Left/Right arrow when focused) to rotate it around its center, or change Gate coverage to resize it. Each edit immediately updates the mask and invalidates the histogram and any peak regions so they can be rerun with the edited gate. Reset fitted gate restores the fitted center, coverage, and rotation.
4. Singlet Gate applies DNA pulse-geometry singlet/doublet discrimination when the required channels exist.

The masked histogram is rebuilt automatically whenever QC changes, and only events within the visible x-axis range are modeled.