Cas9 is a bacterial nuclease guided by a short RNA molecule (the "guide RNA") that base-pairs with one strand of DNA. The Cas9-guide RNA complex slides along the double helix checking each window of DNA against the guide sequence; where the window mismatches the guide in a few positions it may still bind loosely (an "off-target" site), but where it matches almost perfectly it unwinds the helix and cuts both strands, creating a double-strand break the cell then repairs — the basis of gene editing.
P(cut | window) = exp(−mismatches / tolerance)
tolerance = (1 − fidelity) × guideLength / 6
on-target: mismatches = 0 → P ≈ 1
off-target: mismatches ≥ 1 → P falls sharply as fidelity rises
- Scan speed — how fast the complex slides along the DNA searching for a match.
- Guide RNA length — longer guides demand more consecutive matching bases, which statistically shrinks the number of look-alike sequences elsewhere in the genome.
- Nuclease fidelity — engineered high-fidelity Cas9 variants tolerate fewer mismatches before binding collapses, cutting off-target activity at the cost of occasionally missing nothing (the true target still has zero mismatches, so it is always found).
- New sequence — regenerates the DNA strand and re-scatters a few near-match decoy sites so you can watch specificity play out differently each run.
Real-world relevance: this speed/specificity trade-off is exactly what drove the design of high-fidelity Cas9 variants (eSpCas9, HiFi Cas9) used in approved therapies like Casgevy — minimizing off-target cuts is the main safety bottleneck in clinical gene editing.