Flow Cytometry Compute
Most flow cytometry work runs in your browser and uses none of your team's compute usage. Gating, compensation, statistics, plotting, and the cell-cycle and proliferation tools all run on your own computer. Every export is free.
Clustering and dimensionality reduction can also run on Conspecta's servers, for data too large for your computer. A server run uses compute usage. A run left on Local is free.
Choosing Where a Run Happens
Click Clustering or Dim. Reduction in the sidebar of an analysis. The dialog box has a Compute Location setting with two options:
- Local runs in your browser and is free.
- Server sends the run to Conspecta's servers and uses compute usage.
Every run starts on Local, however many events it covers. Nothing goes to a server until you choose Server and start the run.
When you choose Server, the dialog box shows your team's usage so far this month. Under it, Estimated compute usage shows what this run will use.
A server run reads the files saved in your project. It can cover the file you're looking at or the whole working set, which you choose under Run on. A working-set run gives the same answer on a server as in your browser. Flow Advanced Analysis explains working-set runs.
Some runs always stay in your browser:
- Density clustering. The dialog box has no Compute Location setting for it.
- A run on a file that is still uploading. The dialog box shows Runs in your browser where the setting would be, until the upload finishes. For a working-set run, every file in the set has to finish uploading.
What Uses Compute
| Work | Usage |
|---|---|
| PCA, UMAP, t-SNE, K-means, or SOM clustering on a server | 1 per run |
| DBSCAN or Graph clustering on a server | 5 per run |
| Counting populations for an AI assistant | 1 per file |
| Applying a gating strategy for an AI assistant | Free |
The algorithm alone sets what a server run uses. Every algorithm works on a sample of the events, so a five-million-event file uses the same as a fifty-thousand-event one. A working-set run over ten files uses the same as a run on one file. The size of your data and the Subsample setting change how long a run takes, and a lower Subsample finishes sooner at the same usage.
Elsewhere in Conspecta, compute usage goes to AI detection in image analysis.
Limits on a Server Run
- A run reads at most 500,000 events. A single file larger than that is thinned to 500,000 by taking evenly spaced events. A working-set run whose Events per file adds up to more than 500,000 is refused, and the message gives the number per file that fits.
- A run can take up to 30 minutes. If your settings would need longer, the run is refused when you start it, with a note on which settings to reduce.
A refused run uses nothing. If a run is stopped early, because it hit the time limit or the server running it was interrupted, it's marked failed and its usage is returned automatically.
Work an AI Assistant Runs
An AI assistant connected to your project can work on your gating without you opening the analysis. It runs the same code as your browser and gets the same numbers. Connect your AI assistant covers setting one up.
- Counting populations uses 1 per file. One run covers up to 24 files and up to 512 MB of FCS data in total, so two dozen small tubes fit in one run. A run over 512 MB is refused before any usage is charged, and the message gives its total.
- Applying a gating strategy is free and covers up to 96 files at once, a full plate.
- Your assistant can start the same server clustering and dimensionality reduction runs, on one file of a saved analysis or pooled across several of its files. They use the same as when you start them yourself. Reading a finished run back is free.
Your assistant can save a finished clustering run onto the analysis. When you open the analysis later, you see which algorithm ran, on which channels, and what it found, with the option to re-run it and color a plot by its clusters.
Working with Large Files
A run in your browser is limited only by your computer. An analysis can cover a study far bigger than your computer's memory, because a file attached to an analysis but not loaded costs no memory. When you reopen a large analysis, Conspecta loads up to 200 files or 5 GB, whichever comes first, and leaves the rest unloaded. A note says how many it left, and you can load them from Manage files.
Scatter, density, and contour plots draw 50,000 events unless you change Events drawn per plot under Display in the sidebar. Statistics always use every event. Subsample in Clustering and Dim. Reduction is filled in from how many events you have, and if a run in your browser is too slow, lower it first.
Tip: K-means, SOM clustering, and Graph clustering accept up to 64 parameters, so a full high-parameter panel fits in one run. DBSCAN accepts up to eight, because in more dimensions the distances between events stop separating populations reliably.
Checking Your Usage
Go to Settings → Usage to see your team's compute usage. Click Compute Jobs to list recent runs with what each one used. A run whose usage was returned reads Refunded.
Exports you start from Settings → Export, such as Flow Cytometry for your raw FCS files and gating, run on Conspecta's servers. You can follow one in Compute Jobs, where it shows a usage of 0.