Manual vs Automated Gating

Gating is the process of drawing boundaries on flow cytometry plots to identify cell populations. Every flow cytometry experiment requires gating. The question is whether a human draws those boundaries by hand or an algorithm does it.

This page compares the two approaches. It explains why manual gating creates reproducibility problems and why modern high-dimensional panels make it impractical.

What is gating?

Flow cytometry produces measurements for each cell that passes through the instrument. To identify specific cell types, you draw regions on 2D scatter plots. Each region selects cells that fall within a defined range of marker expression. These regions are called gates.

Gates are applied sequentially. You start with broad gates to remove debris and doublets. Then you progressively narrow the population with marker-specific gates. The result is a hierarchy of gates that identifies specific cell types.

Standard PBMC gating hierarchy

When analyzing peripheral blood mononuclear cells, most labs follow the same gating strategy.

Step 1: Remove doublets. Plot FSC-A vs FSC-H. Singlets fall along the diagonal. Doublets have higher area relative to height and fall off the diagonal. Draw a gate around the diagonal to keep only singlets.

Step 2: Identify lymphocytes. Plot FSC-A vs SSC-A. Lymphocytes are small with low granularity. They appear as a tight cluster in the lower left. Monocytes are larger and slightly more granular. Granulocytes are large and highly granular.

Step 3: Separate T cells and B cells. Gate on the lymphocyte population. Plot CD3 vs CD19. CD3+ cells are T cells. CD19+ cells are B cells.

Step 4: Separate CD4 and CD8 T cells. Gate on CD3+ cells. Plot CD4 vs CD8. You get four quadrants: CD4+CD8- (helper T cells), CD4-CD8+ (cytotoxic T cells), double positive, and double negative.

Tip

FMO (Fluorescence Minus One) controls are essential for setting gate boundaries. An FMO control contains all antibodies in the panel except one. The spread you see in the missing channel tells you where to draw the gate boundary. Without FMOs, you are guessing where positive starts and negative ends.

The reproducibility crisis in manual gating

Manual gating looks straightforward for a 4-color panel. Draw some polygons, count the cells inside. The problem is that different people draw different gates on the same data.

Inter-operator variability

Studies have measured what happens when multiple analysts gate the same FCS files. The results are concerning. Inter-operator variability of 10 to 30% on population frequencies is common. One analyst gates 12% CD4+ T cells. Another gates 16% on the same sample. The cells did not change. The gate placement did.

This variability comes from subjective decisions. Where exactly does the lymphocyte cloud end? How tight should the singlet gate be? Where is the boundary between CD4-positive and CD4-negative? These decisions vary from person to person and even from day to day for the same analyst.

No standard reporting

Publications rarely report gating strategies in enough detail to reproduce the analysis. The MIFlowCyt (Minimum Information about a Flow Cytometry Experiment) reporting guidelines exist to address this. They specify that gating strategies should be fully described and that representative gating plots should be included. In practice, these guidelines are rarely followed.

FlowCAP benchmarks

The FlowCAP (Flow Cytometry: Critical Assessment of Population Identification Methods) challenges compared manual and automated gating approaches. A key finding was that automated methods can match or exceed the consistency of manual gating. When multiple automated methods are applied to the same data, they tend to agree more with each other than multiple human analysts do.

The dimensionality problem

The real limitation of manual gating is not just reproducibility. It is dimensionality.

Traditional flow cytometry: 4 to 8 colors

With a small panel, manual gating with bivariate plots is feasible. You have a manageable number of 2D plot combinations. An experienced analyst can gate a 6-color panel in a few minutes.

Spectral flow cytometry: 20 to 40+ parameters

Modern spectral flow cytometers can resolve 20 to 40 or more fluorochromes simultaneously. Manual gating with bivariate plots breaks down at this scale. The math is simple: with N markers, you need N*(N-1)/2 bivariate plots to examine all possible pairs. A 30-parameter panel produces 435 possible bivariate plots.

No analyst can systematically examine 435 scatter plots per sample. Populations defined by combinations of many markers become invisible when you can only look at two dimensions at a time.

Mass cytometry (CyTOF): 40+ parameters

Mass cytometry uses metal-tagged antibodies instead of fluorochromes. There is no spectral overlap because each metal isotope has a distinct mass. Panels of 40 or more markers are routine.

With 40 markers, you would need 780 bivariate plots. Manual gating is essentially impossible for full exploration of the data. Most CyTOF studies use automated approaches for population identification.

Populations hide in high dimensions

Consider regulatory T cells (Tregs). They are defined as CD3+CD4+CD25+CD127low. That is four markers. A manual gating hierarchy can find them. But what about a subset of Tregs that also express specific chemokine receptors, activation markers, and transcription factors? A population defined by 8 or 10 markers simultaneously cannot be found by looking at 2D plots one pair at a time.

High-dimensional data contains populations that only become visible when you consider many markers together. This is where automated methods excel.

Automated gating approaches

Unsupervised clustering

Unsupervised methods analyze all markers simultaneously. They group cells into clusters based on their full expression profiles.

FlowSOM uses self-organizing maps to cluster cells. It first maps cells onto a grid, then performs meta-clustering to group similar grid nodes. FlowSOM is fast and scales well to millions of cells. It is the most widely used unsupervised method for flow and mass cytometry.

PhenoGraph builds a nearest-neighbor graph and applies Louvain community detection. It tends to find more granular populations than FlowSOM. PhenoGraph works well for discovery but can be slow on very large datasets.

Leiden clustering is similar to Louvain but produces better-connected communities. It is increasingly used in cytometry, especially through the scanpy ecosystem.

All of these methods find populations across all dimensions simultaneously. They can identify cell types that manual gating would miss.

Template-based automated gating: openCyto

openCyto takes a different approach. Instead of unsupervised clustering, it automates the traditional gating hierarchy. You define your gating strategy in a CSV template that specifies which markers to gate, in what order, and what algorithm to use for each gate.

The result is a reproducible gating hierarchy that mirrors manual gating but removes the subjective element. Every sample is gated with exactly the same rules.

Advantages of automated gating

Aspect	Manual	Automated
Reproducibility	Low. Different analysts, different results.	High. Same algorithm, same result every time.
Dimensionality	Limited to 2D plots. Practical up to ~8 markers.	Handles all dimensions simultaneously. No practical limit.
Speed	Minutes to hours per sample.	Seconds to minutes per sample.
Novel populations	Only finds what you look for.	Can discover unexpected populations.
Domain knowledge	Requires deep expertise in marker biology.	Requires understanding of algorithm parameters.

Disadvantages of automated gating

Automated methods are not perfect. They require validation against known biology. A clustering algorithm does not know that CD3+CD4+ cells are helper T cells. It just knows they form a cluster. You still need biological expertise to interpret the results.

Algorithms can also split real populations into multiple clusters. Or they can merge distinct populations into one cluster. The results depend on hyperparameters like the number of clusters, the number of nearest neighbors, or the resolution parameter. Different parameter choices can give different results.

Caution

Never trust automated gating results without validation. Always check that known cell types are correctly identified. Overlay cluster assignments on traditional bivariate plots to verify that the algorithm found biologically meaningful populations.

The hybrid approach

Most experienced cytometrists use a combination of both methods. The practical workflow looks like this:

1. Manual gating for cleanup. Use manual gates for singlets, live cells, and the initial lineage gates. These are well-defined populations with clear boundaries.
2. Unsupervised clustering for phenotyping. Apply FlowSOM or PhenoGraph to the cleaned population. Let the algorithm find subpopulations across all markers.
3. Validation. Check that the clusters correspond to known biology. Examine marker expression heatmaps. Overlay clusters on UMAP or tSNE plots.

This hybrid approach combines the reliability of manual gating for well-known populations with the power of automated methods for high-dimensional discovery.

Key takeaways

Manual gating works for small panels and well-characterized populations. It is intuitive and easy to understand. But it suffers from reproducibility problems and cannot handle modern high-dimensional panels.

Automated gating is reproducible, fast, and handles any number of markers. It can find populations that manual gating misses. But it requires careful parameter tuning and biological validation.

For panels with more than 10 to 15 markers, automated approaches are not just helpful. They are necessary. The data contains information that bivariate plots simply cannot show.