Review a CRISPR target

Ask: "Using rnafold, find SpCas9 guides in this sequence and flag the ones with problems." The example is the first 200 nucleotides of the avGFP coding sequence.

The assistant calls find_crispr_guides, which scans both strands and returns every candidate with its position, strand, PAM, GC fraction and any flags. Real output (examples/crispr_guides.json): 19 candidate guides.

The first one:

spacer  AAAGGAGAAGAACTTTTCAC   PAM TGG   strand +   positions 10-29   GC 0.35
flag    TTTT at spacer position 14: a run of four or more T terminates RNA polymerase III,
        so a U6-driven guide would be cut short

A flag names a specific, checkable property and its threshold; it is not a verdict on the guide.

What it deliberately does not check

The tool lists this in the result (not_checked), and your assistant should repeat it:

  • Off-target sites anywhere in any genome. No genome was consulted, and there is no validated off-target model.
  • Chromatin accessibility, methylation, or anything about the cellular context.
  • On-target efficiency. The guides are not ranked: the order is positional, not preferential.

So the output answers "which spacers are structurally plausible and free of obvious problems", and not "which should I use". Take your shortlist to a genome-aware off-target tool before ordering anything.

Reverse-strand guides are reported in your own numbering.

Tools used

find_crispr_guides; optionally scan_vector_liabilities and find_restriction_sites on the construct (free).

Last changed 2026-11-17