Batch Editing Quantification — A Whole Plate of Edited Wells Against One Control

They are both plate-scale and they answer different questions. Sanger Clone Screening judges CLONING: it groups a plate of reads into clones and gives each one a verdict — PASS, point mutation, indel, vector only, wrong insert, low coverage, ambiguous — so you know which colony to grow up. This page measures EDITING: each well is a pool of cells, not a clone, and what comes back is a percentage of that pool — how much of it carries an indel, the base conversion, or the intended knock-in. If you are picking colonies, use the screening page; if you are ranking guides, editors or donor concentrations, use this one.

Indel mode decomposes each well onto shifted copies of the control and reports the edited fraction and the dominant indel size. Base mode reports the conversion at the target base against that well's own background, with the bystander summary and noise floor. Knock-in mode fits the intended edit against wild type and unintended indels. One call runs one mode over the whole plate, and each mode is blind to what the others measure — indel mode cannot see a base edit (a substitution makes a mixed base, not a shift), base mode cannot see an indel — so a plate that had both done to it is two runs over the same files.

The fit badge is the single-sample tool's OWN gate verdict on that row, called directly rather than re-implemented here, so the plate can never say a well is fine while the per-well tool says it is not. It is not a small number: a percentage from a well whose model fitted badly is a different kind of statement from one that fitted, so the two must not be ranked against each other. You can still sort every column, and the rows that failed the fit are marked rather than hidden — but the plate summary (min, median, mean, max) is computed over the adequate rows only, and says how many wells that left out.

It becomes a failed ROW carrying its own error message, and the rest of the plate still comes back — that isolation is the whole reason a plate tool exists, since a batch that aborts on well 37 is useless to someone holding 96 wells. What is NOT isolated is a problem that is wrong for every row: a control trace that cannot support the mode, an argument belonging to another mode, an out-of-range limit or a plate over the work ceilings all raise one error instead, because returning 96 identical failed rows would dress a single mistake up as a plate-wide biological result.

No validation is published for this implementation, and being a batch does not soften that: every row is exactly the claim the corresponding single-sample tool makes — a composition of a real pool, checkable only by amplicon sequencing. There are no fitted coefficients anywhere; the basis (or the background) is your own control trace, and the models are TIDE's shifted-control decomposition for indels (Brinkman et al., Nucleic Acids Research 2014), TIDER's added intended-allele column for a knock-in, and EditR's two-peak mixture for base editing (Kluesner et al., The CRISPR Journal 2018). What each row reports instead is what it rests on — R² for indel and knock-in mode, the background n, sd and noise floor for base mode — plus the fit verdict. Recovery of synthetic mixtures is deliberately not offered as validation, because it tests the arithmetic and the plate plumbing rather than whether the model fits a real capillary trace.

Up to 384 samples per call — 96 is the natural plate — with at most 3,000 called bases per trace and 400,000 trace positions across the whole call, control included. The position ceiling is the one that usually binds, because every channel row is four numbers and 400,000 of them is already an 8-10 MB request. A plate of untrimmed reads is told to split rather than left to fail somewhere in transit, and splitting changes no well's number: every call re-reads the same control.

They are suffixed, never merged, and the second suffix is checked against every id on the plate so it can never land on another well's real name. Both rows survive with distinct keys and the gate reports that it happened, because merging two wells would average two different samples into one confident row. In practice a duplicate id means the plate map is off by a row or a column, so fix the ids on the page before reading the table.

An alignment offset is a property of ONE pair of reads — where that well's read happened to start — so a single value imposed on a plate would be right for at most one well and would misregister the rest. Every row finds and reports its own offset instead. Sending one anyway is rejected rather than ignored, since an argument that is accepted and silently dropped changes the answer and says nothing; if you genuinely need to force an offset for one well, quantify that well with the single-sample tool.