Filter set and channel checker
Check a multi-colour panel against your filters before you stain anything.
Everything that will be on the slide, including counterstains.
Channels
Where each fluorophore’s signal goes
| Fluorophore | Green | Red | Excited | Collected |
|---|---|---|---|---|
| EGFP | 100.0% | — | 100% | 64% |
| mCherry | — | 100.0% | 64% | 42% |
Each row is one fluorophore as a percentage of its own best channel, so it does not depend on how much is expressed or on how bright the fluorophore is — those cancel. The last two columns are, in that best channel, the fraction of peak absorptivity the illumination reaches and the fraction of emitted photons the filters pass.
What each channel is looking at
- Green — 100.0% EGFP
- Red — 99.8% mCherry
This one assumes equal molar amounts of every fluorophore, which is almost never true — a strong promoter and a knock-in tag differ by orders of magnitude, and that swamps everything here. Read it as a ranking, not a prediction. The table above carries no such assumption.
collected = ∫EM(λ)·T(λ)dλ ÷ ∫EM(λ)dλ; excited = EX(λ_laser), or ∫EX·T dλ ÷ ∫T dλ through a filter- Fluorophore spectra, extinction coefficients and quantum yields — FPbase, Nature Methods, 2019
- Bleed-through, filter choice and the controls a multi-colour experiment needs — Journal of Cell Biology, 2006
Calculated in your browser. Nothing you entered was sent anywhere.
Filters are modelled from their designation, not from a measured curve, so any part you can name works — but out-of-band blocking is taken as perfect and a real filter is not. Where a channel uses an excitation filter rather than a laser, the lamp is assumed flat across the band; LED and metal-halide sources are not, and a source spectrum would change the excitation column by a few per cent. Detector sensitivity is not modelled at all, which matters past about 700 nm on a silicon camera.
When to use this
Use this before staining anything, when you have a panel in mind and a microscope with particular filters, to find out what each channel will really collect. It is also the tool for diagnosing a channel that looks contaminated after the fact. It models filters from their designation rather than measured curves, so any part you can name works; it does not model your detector, which starts to matter past about 700 nm.
Worked example
EGFP and mCherry on a confocal: a green channel at 488 with a 525/50 emission filter, and a red channel at 561 with 600/50.
- Fluorophores
- EGFP, mCherry
- Green channel
- 488 nm, 495 LP, 525/50
- Red channel
- 561 nm, 570 LP, 600/50
Result
The panel works, and the interesting number is the last one: the red channel collects only 42% of what mCherry emits, so a wider emission filter is the cheapest signal available here — cross-talk is not what is limiting this experiment.
What people get wrong
- Choosing an emission filter from the fluorophore alone and never checking it against the other labels. A 600/50 is a perfectly good mCherry filter and a poor one if there is a TagRFP in the next channel, and nothing about the single-colour view reveals that.
- Blaming the filters for a fluorophore the illumination barely reaches. If a label is only 5% excited in its own best channel, no emission filter recovers light that was never emitted — that needs a different laser line, and the tool says so rather than suggesting a filter change.
- Forgetting that sequential excitation removes most cross-talk for free. DAPI emits well into a green passband, but at 488 nm it absorbs nothing at all, so a sequentially acquired green channel sees exactly none of it. The same two filters with a shared violet line leak badly.
- Reading the channel composition figures as a prediction. They assume equal molar amounts of every fluorophore, and a strong promoter against a knock-in tag differs by orders of magnitude — enough to swamp the entire calculation. The bleed-through table carries no such assumption.
Questions
+How do I write my filters?
As they are printed: a bandpass as centre/width such as 525/50, and an edge filter with LP or SP such as 495 LP. Vendor part numbers including ET525/50m and FF01-525/50-25 are read correctly. A bare number is refused, because 525 alone could be either a bandpass centre or a longpass edge.
+Why is bleed-through independent of expression level?
Because each row compares one fluorophore against itself in two channels. How much of it is present, and how bright it is, multiply both numbers equally and cancel out. That is what makes the figure worth trusting when nothing else about the sample is known.
+What does "fit to" do to a channel?
It derives a filter set from that fluorophore’s own spectra: the dichroic goes where its excitation and emission curves cross, the excitation band runs from there out to half maximum, and the emission band out to a fifth of maximum. It knows nothing of any vendor catalogue, which is why agreeing closely with the standard cubes is worth something.
+How accurate is modelling a filter instead of measuring it?
Good enough for the question being asked. The model assumes perfect out-of-band blocking, and a real hard-coated filter blocks to about one part in 100,000 — far below the in-band spectral overlap that actually causes cross-talk, which is modelled properly. It will understate leakage from a damaged or badly angled filter.
Related tools
- Fluorescence spectra viewer — Overlay excitation and emission spectra and see which pairs will separate.
- Fluorophore brightness comparison — Rank fluorophores by what your setup will actually detect, not by ε × Φ alone.
- FRET pair calculator — Förster radius from real spectra, and the artefacts that will spoil the measurement.
Science last reviewed .