CRISPR Off-Target Checker — CFD-Scored Guide Screen
Screen a guide's protospacer against curated lab reference genomes — SpCas9 hits ranked by CFD, not by mismatch count.
🌐 Nothing you paste is logged or stored — every tool is also callable via REST & MCP, and in bulk from the batch tools
Predicted, not measured
- How good is it?
- Best of the common off-target scores on the authors' GUIDE-Seq comparison, at Pearson r = 0.40 over 9 guides and 402 sites (vs CCTop 0.31, Hsu-Zhang 0.26) — a useful ranking, not a reliable magnitude. Weights were measured for SINGLE mismatches; multiple mismatches are multiplied, and Listgarten et al. 2018 note the training data never contained a mismatch and an alternative PAM together, so that combination is extrapolation.
- Only valid for:
- SpCas9 with a 20 nt spacer, which is the only case scored — every other nuclease returns null rather than a number from the wrong enzyme's table. One locus, one cell line. Substitutions only: DNA/RNA bulges are neither searched nor scorable, and a low score is not a claim that a site is safe.
- Fitted on:
- Doench et al. 2016 CFD weights (Nat Biotechnol 34:184): the percent of on-target activity retained by each single mismatch at each of the 20 positions, and by each PAM, measured in one pooled screen tiling human CD33 in MOLM13 cells (27,897 sgRNAs; 65 active perfect matches and 9,914 variants).
Paste a guide's protospacer and screen it for off-target sites — a protospacer match plus a valid PAM, on both strands — across a curated set of common lab reference genomes. For SpCas9 with a 20 nt spacer each site is also scored with Doench 2016 CFD, so the list is ordered by predicted cut likelihood rather than by mismatch count: two sites carrying the same number of mismatches can score several-fold apart, because where a mismatch falls matters far more than how many there are. Every site is shown aligned against your guide with the differing bases highlighted, so that difference is visible rather than asserted. The search covers the bundled genomes only — not the human, mouse or rat nuclear genome — so a guide headed for mammalian cells still needs a genome-wide screen elsewhere.
A small curated set of common lab reference genomes — bacterial chromosomes plus the human mitochondrion, roughly 9.7 Mbp in total. Not the human, mouse or rat nuclear genome. For mammalian guide design this is a sanity check on the host and vector side and nothing more; the genome-wide search still has to happen somewhere else. Every run lists the exact genomes it covered, and anything absent from that list was not looked at.
0 nt after cleaning. SpCas9 uses a 20 nt spacer.
Working on a whole construct rather than one guide? Open SeqStudio — the full editor, with live feature annotation, plasmid maps, restriction and primer panels, undo/redo, multi-document tabs and GenBank / SnapGene import and export.
How to use the CRISPR Off-Target Check tool
- 1Paste the protospacer on its own — the spacer you would clone, 5′→3′, with no PAM and no flanking template.
- 2Pick the nuclease (its PAM is what turns a sequence match into an actual site) and how many mismatches to tolerate, up to 4.
- 3Read the ranked sites: each shows its genome and coordinates, its mismatch count, its CFD score, and the off-target protospacer aligned against your guide with the mismatched bases picked out.
- 4Check the list of genomes searched and the gate's "not checked" notes before concluding anything — a quiet result covers those genomes at that mismatch budget, and nothing else.
Frequently asked questions
Which genomes does this actually search?
A small curated set bundled with SeqBench — Escherichia coli K-12 MG1655, Bacillus subtilis 168, Mycoplasma hyorhinis SK76 and the human mitochondrion (rCRS), about 9.7 Mbp in total — and every run lists exactly what it covered. This is not a human, mouse or rat nuclear genome search. What it catches is off-target cutting in the host you are cloning in, in a common cell-culture contaminant, and in mitochondrial DNA; a guide destined for a mammalian nuclear target still needs a genome-wide screen somewhere else.
What is CFD, and why rank by it instead of by mismatch count?
CFD (Cutting Frequency Determination, Doench et al. 2016) estimates the fraction of a guide's on-target activity that survives a given off-target's mismatches and PAM, using weights measured one position at a time. Position dominates: a mismatch next to the PAM costs far more activity than the same mismatch at the far end of the spacer. That is why two sites with an identical mismatch count can land several-fold apart, and why a ranking by count alone puts genuinely different sites side by side.
What does the CFD specificity number mean?
It is 1 / (1 + the sum of CFD over the off-target sites found), the GuideScan aggregate, on a 0–1 scale where lower means more predicted off-target load. Exact matches are excluded from the sum, since a perfect protospacer+PAM match in the organism you are targeting is the on-target. Crucially it scales with how hard you looked — a tighter mismatch budget or a smaller genome set both push it up — so it is only comparable between guides screened with identical settings.
Does a low CFD score mean a site is safe?
No. CFD ranks sites; it does not clear one. The weights come from one pooled screen at one locus in one cell line, single mismatches were measured and multiple mismatches are multiplied, and on the authors' own GUIDE-Seq comparison the correlation with measured off-target cutting was r = 0.40 across 9 guides. Treat the score as a shortlist order for what to sequence, not as a verdict.
Which nucleases can be scored?
SpCas9, SpCas9-NG, SaCas9 and Cas12a are all searched, each with its own PAM and PAM side. Only SpCas9 with a 20 nt spacer is CFD-scored, because that is the only case the weights were measured on; anything else returns "not scored" rather than a number borrowed from the wrong enzyme's table, and no aggregate specificity is shown.
Does it find off-targets with insertions or deletions?
No — substitutions only, up to the mismatch budget you set (4 maximum). DNA and RNA bulges are neither searched nor scorable here, since CFD requires the guide and the site to be the same length. Bulged off-targets are real, so their absence from the list is not evidence they do not exist.
Is my sequence stored, and can I use this from code?
No. Your protospacer is used only to run the screen and is never logged or stored. The crispr_offtarget_check tool is available via the REST API and MCP server, and in bulk from the batch tools.
More
Related tools
Convert gene symbols, Ensembl, Entrez GeneID, RefSeq and UniProt accessions in batch, and find cross-species orthologs.
Design SpCas9 pegRNAs for any substitution, insertion, deletion or small replacement — spacer, PBS sweep, RTT and 3' extension, PE3 nicking guides, plus twinPE dual-pegRNA design for large edits.
Score every PBS/RTT combination for one prime edit with PRIDICT2.0 and rank the pegRNAs — HEK293 and K562 scores, library percentiles, the full pegRNA, and Golden Gate cloning oligos.