The Mistakes We Keep Making in Flow Cytometry, and Why They're Not Really Mistakes

The following article is an opinion piece written by Rosa Sottile, PhD. The views and opinions expressed in this article are those of the author and do not necessarily reflect the official position of Technology Networks.

After many years working in flow cytometry, first as a user myself running complex immunological experiments and now helping other users in an academic core, I’ve concluded that most of the mistakes we keep seeing in this field are not really mistakes.

They are a recurring and predictable consequence of how flow cytometry knowledge has historically been transferred across labs, papers, workflows, and training environments, often without expert training from an experienced flow core. I observe the same errors showing up in panel after panel, lab after lab. When something is this consistent, we can’t blame just the individual users anymore—and the pattern becomes the message.

If we want to fix how we do flow, we need to address the gaps in the way we teach and learn the technique by rethinking when and how concepts are introduced, received, and applied during training. Addressing these gaps can lead to significant improvements in data quality and reproducibility of flow experiments. Here are four of the most common mistakes I see, along with what they tell us about the gaps in our teaching, training, and learning culture.

1. Designing panels before defining the biology

Unlike cloning experiments, where failure is obvious, you can still run an apparently clean flow cytometry panel and think of it as a technical success; while the truth is that biological failure has gone undetected by an inexperienced eye.

Many users start an experiment with a list of markers, often copied from a paper, inherited from a former lab member, or pieced together from a list of antibodies already present in the fridge. What users often lack is a clear structure of the populations they need to resolve, the level of resolution required, or the downstream comparisons the data will support. Without a clear biological vision, panel design becomes a shopping exercise, rather than a true experimental design.

This reflects how users approach flow cytometry: first instrument setup, then acquisition and marker selection. The concept of experimental design often matures later with experience, troubleshooting, and mentorship. Trainees learn to use the cytometer before learning how to think with it. Panel design should begin with a clear biological question and a defined cell hierarchy, with a list of populations the experiment must resolve. Fluorochrome assignment should come last, not first.

2. Treating controls as optional, or interchangeable

In flow cytometry, controls are very often skipped, compromised, or substituted. Unstained samples are run on a different cell type, and single-stain controls often don’t match the experimental tubes. Fluorescence-minus-one (FMO) controls are treated as optional rather than necessary, particularly when samples are precious or limited.

These behaviors individually feel small, but especially with large panels, they can compromise the entire interpretation of the experiment. Without a proper unstained control, autofluorescence can go unnoticed. Without proper single-stain controls, compensation or unmixing is unreliable. Without FMOs, gating populations based on continuous markers is essentially a matter of guesswork.

Controls keep getting skipped because they are introduced to users as a separate topic, after the panel is already built, rather than as a structural checklist that is part of the experimental planning. They should be designed and budgeted at the same time as the panel and treated as non-negotiable. If this is not happening, experiments become far more vulnerable to interpretive errors.

Sample preparation is the invisible partner to this and deserves its own piece. Fixation, dissociation, buffer choices, and how peripheral blood mononuclear cells (PBMCs) or whole blood are processed all shape the data as much as any antibody choice does.

3. Skipping titration

Antibody titration is the single most cost-effective quality control step users can implement, and yet, is one of the most consistently skipped good practices. The dominant idea is still to use the manufacturer’s recommended concentration, on the assumption that whoever is selling the antibody knows best.

This assumption is wrong and fails in practice. These recommended concentrations are derived under one specific set of conditions, which rarely match the user's actual protocol. If an antibody is used at a very high concentration, it can increase non-specific binding, leading to a higher background and compromising the resolution or overall performance of the panel. For larger panels, the impact on performance is even more consequential, as it can lead to a higher spread in neighboring detectors. If used at a suboptimal concentration, dim populations cannot be resolved properly. Either way, biology suffers. 

Titration is usually skipped because it is often treated as an advanced topic, underestimated, or seen as a step that delays data collection. In many research environments, practical pressure naturally prioritizes rapid data generation over optimization steps. Thinking of titration as an essential step to obtaining high-quality data is one of the highest leverage changes we can make.

Improving Oral Bioavailability Prediction With Human Models

Join this webinar to explore how an integrated human gut/liver microphysiological system (MPS), combined with mechanistic modeling, provides deeper insight into oral drug disposition.
View Webinar

Advertisement

4. Treating compensation or unmixing as software issues

Compensation in conventional flow and unmixing in spectral flow are mathematical operations calculated by the software. But reducing them to purely software factors can be misleading. Both operations depend entirely on the quality of the single-stain controls acquired in the instrument, but even before that, on the user’s understanding of what the algorithm is doing in the background.

In practice, compensation or unmixing is treated as a single-click option after single-stain controls are run quickly and without verification. The resulting matrices are accepted without inspection, or even worse, manually modified post-acquisition. In spectral cytometry, the handling of single-stain controls matters even more, because subtle differences in their optical properties—from tandem degradation, lot-to-lot variability, or mismatched vials—introduce persistent unmixing errors that propagate through every downstream gate.

Compensation and unmixing are usually taught as workflow steps rather than as concepts that must be understood before any experiment is performed. Users often learn the order of clicks, but not what spillover is, why a positive control must be at least as bright or brighter than the sample, or why autofluorescence unmixing is sometimes not optional in spectral mode. Without that conceptual layer, the software cannot save any experiment, because software cannot tell when it is producing nonsense.

The pattern, and what to do about it

The four mistakes above look unrelated on the surface. But they share a common structure: each happens when a conceptual step is replaced by a procedural one. Defining the biology becomes picking markers. Designing the controls becomes running them if there's time. Titrating becomes using the recommended concentration. Understanding spillover becomes clicking the compensation button.

These patterns are not surprising when conceptual understanding and procedural training develop separately. Under those conditions, errors perpetuate across laboratories and experience levels.

The good news is that this is fixable without new instruments, new reagents, or new software. It requires only the integration of experimental design principles earlier into scientific workflows: biology before panel, controls alongside panel, titration as baseline, and keeping algorithms as concepts rather than buttons. Until that reordering becomes the default, we will keep producing flow data that is technically generated, but conceptually fragile. In flow cytometry, garbage in is garbage out, and the inputs start with the training.