SeqBench

CRISPR Editing Efficiency from Sanger — Indel Spectrum Without NGS

Edited trace decomposition against an unedited control: indel spectrum, unedited fraction, R².

Your files are never stored or shared

Predicted, not measured
How good is it?
Every run reports its own R², which says how well this model fits YOUR two traces — a low value means the assumption is wrong for these reads, not that the edit is weak, and the result should not be used. There is no published accuracy for this implementation against amplicon sequencing, so treat the percentages as an estimate to confirm if a number matters.
Only valid for:
A pool of alleles that differ from one control read by simple insertions or deletions at a known cut site, where both reads come from the same amplicon and chemistry and both extend well past the cut. NOT valid for substitution-only edits — base editing produces a mixed base, not a shift, and this model cannot see it — nor for a knock-in whose insert is novel sequence rather than a frame shift of the control, nor for any run whose R² comes back low.

Quantify CRISPR editing efficiency from two Sanger traces, without amplicon sequencing. A Sanger trace of an edited pool superposes every allele in it: downstream of the cut, an allele carrying an indel of d bases reads as the unedited sequence shifted by d, so the trace over a window past the break is a combination of shifted copies of the control trace, and the mixing proportions are the indel spectrum. The decomposition runs in your browser. Every run reports its own R², how much of the window the shifted-control basis actually explains, because that rather than the efficiency number is what tells you whether the model fits your traces at all: a low value means the assumption is wrong, not that the edit was weak. One limit before you start: a base editor makes a mixed base rather than a shift, so a CBE or ABE experiment reads as unedited with a confident-looking fit. This is for nuclease editing.

Choose fileor drop it here
Choose fileor drop it here

Still needed: the unedited control trace and the edited pool's trace.

Design, edit, and verify complete constructsIt opens on the CRISPR designers, against this sequence.

Base editing instead? CBE and ABE rates from the same two traces

A base editor changes a base rather than shifting the register, so an indel spectrum does not model it. Each editable position in the activity window is treated as a mixture of the two bases, and the per-position conversion is reported, bystander positions included.

Choose fileor drop it here
Choose fileor drop it here

Both traces must be the same amplicon on the same chemistry; the control is what the background is subtracted from.

In standard deviations of this sample's own background. Raising it raises the reported noise floor with it.

Still needed: the untreated control trace, the edited pool's trace and a protospacer of at least 12 nt.

Knocking a specific allele in? Measure that allele, not a ladder

Where the intended sequence is known, decomposition onto a generic indel ladder discards that information. The intended allele is placed in the basis alongside wild-type and the indel background, so the reported rate is that of the intended edit.

Choose fileor drop it here
Choose fileor drop it here

Still needed: the untreated control trace and the edited pool's trace.

A whole plate against one control

The same three analyses at plate scale: one untreated control trace against 24–96 edited wells, in whichever of the three modes above applies, returned as a single sortable table.

Indel mode does not detect substitutions and base mode does not detect indels; a plate edited both ways needs two runs.

Choose fileor drop it here
Choose filesor drop them here

The control must share its amplicon, primer and chemistry with every well.

Load the control trace and the cut is located for you.

▸Deconvolution settings: max indel 10 bp · window 70 · guard 3

Required only when the protospacer above does not locate.

Max indel 1–20 bp; window 20–300 bases downstream of the cut; guard = bases skipped immediately after it. A value outside its range is rejected rather than moved to the nearest legal one: a substituted limit would alter every row identically and leave nothing in the table looking anomalous.

Still needed: the untreated control trace and at least one edited well.

How to use the CRISPR Editing Efficiency tool

  1. 1Load the unedited control trace and the edited pool's trace, both .ab1 — or click Load example to see the decomposition on a synthetic mixture.
  2. 2Give the cut site: paste the 20 nt protospacer to have it located in the read, or type the position directly.
  3. 3Read the spectrum, the fraction of the pool at each insertion or deletion size, against the unedited fraction the efficiency is measured from.
  4. 4Check the R² before relying on any of it, and widen maxIndel or the window if the fit is poor.

FAQ

How does this work without sequencing?

A Sanger trace of a mixed pool already contains every allele, superposed rather than separated. Past the cut, an allele with an indel of d bases is the unedited sequence shifted by d, so the observed trace is a weighted sum of shifted copies of the control trace and the weights are the spectrum. Recovering those weights has one exact answer, so nothing is fitted or tuned; this is the same decomposition TIDE performs.

What does the R² mean, and why is it shown so prominently?

It is how much of the variance in the decomposition window the shifted-control basis explains, that is, how well the model fits your two traces. It is a measurement rather than a claim: below about 0.7, read the result as “these traces do not fit this model” rather than as a weak edit. A failed trace, the wrong amplicon, a mistaken cut position and a substitution-only edit all land there.

Will it work for base editing?

No, and it will not announce the failure. A cytosine or adenine base editor produces a mixed base at one position, not a frame shift, so it contributes nothing to any shifted column: the decomposition returns roughly zero per cent edited with a high R², which looks like a confident negative result. Use this for nuclease editing that makes indels. For a base edit, compare the traces position by position instead; the Sanger vs Reference tool shows the mixed peak directly.

How accurate is it?

No accuracy figure against amplicon sequencing is published for this implementation, and none is quoted. Two things are offered instead. The first is the per-run R² above, a measured property of your own data. The second is a limit rather than a strength: at a few per cent channel noise, a genuinely unedited sample still decomposes to a small non-zero edited fraction, so a result of one or two per cent should be read as the method's floor rather than as a real edit. This implementation does recover known synthetic mixtures essentially exactly, but that tests the arithmetic, not whether the model fits a real capillary trace, so it is deliberately not offered as validation.

Why does it need a control trace at all?

Because the control is the model: nothing is fitted, and the answer is built entirely out of your own unedited read, on the exact amplicon, chemistry and instrument you used. Nothing is transferred from another dataset, but the control has to be the same amplicon, run the same way, and reach well past the cut. The result reports the base-call identity it found upstream of the cut, so that can be checked.

Which specific indels does it name?

It resolves indel length, not sequence: two different three-base deletions are one column of the decomposition and are reported together. Both alleles shift the downstream trace by the same amount, so no arithmetic separates them. For the actual junction sequences, use amplicon sequencing.

Method and validation

Model:
Mechanistic — a physical model with fitted constants.
Fitted on:
Nothing. There are no fitted coefficients anywhere in this tool: the basis is the caller's own control trace, and the only free parameters are the mixing proportions themselves, solved per run under non-negativity and a sum-to-one constraint. The model — that an allele carrying an indel of d bases reads downstream of the cut as the unedited sequence shifted by d — is the decomposition TIDE introduced (Brinkman et al., Nucleic Acids Research 2014).
Validation:
Every run reports its own R²: how much of the observed window the shifted-control basis actually explains. That is a measured adequacy of the model on YOUR traces, and a low value means the assumption is wrong here rather than that the edit is weak. On accuracy against a reference method for real samples, none is published for this implementation — the underlying decomposition is TIDE's, whose authors report their own concordance with amplicon sequencing, and that number does not transfer to this code so it is not quoted. This implementation's exact recovery of synthetic mixtures is deliberately not offered as validation either: it tests the arithmetic, not whether the model fits a real trace.
Helpful?