CRISPR guide RNA (sgRNA) design in silico: a practical workflow

What an sgRNA needs

A CRISPR guide RNA is a short RNA (commonly a 20-nucleotide spacer for SpCas9) that directs the Cas nuclease to a matching genomic site. The site must sit next to a protospacer adjacent motif (PAM): for SpCas9 the PAM is NGG, immediately 3′ of the target. A good guide is specific (few close off-targets), active (cuts efficiently), and positioned correctly for the edit you want.

Step 1: Choose the target site for the edit

Where the guide should cut depends on the goal. For a knockout, target an early, constitutive exon so a frameshift disrupts the protein, and avoid the very first exon and alternative start sites. For a knock-in or a precise edit, position the cut as close as possible to the intended change, because homology-directed repair (HDR) efficiency falls quickly with distance from the cut. For CRISPRa or CRISPRi, target the promoter or transcription start site rather than the coding sequence.

Step 2: Find candidate guides next to a PAM

Scan both strands of the target region for 20-nucleotide sequences immediately upstream of an NGG PAM (for SpCas9). Other nucleases use other PAMs (for example SaCas9 or Cas12a), so match the PAM rule to the enzyme you will use. This usually yields many candidates, which the next two steps rank.

Step 3: Score on-target activity

Not all guides cut equally well. On-target activity is predicted from the guide sequence using empirically trained models, most commonly the Doench Rule Set 2 (also called Azimuth), which was trained on thousands of guides from genome-wide screens. Prefer guides with high predicted activity, and as a rough rule avoid extreme GC content and long single-base runs (for example four or more T’s, which can terminate transcription of the guide).

Step 4: Predict and minimize off-targets

Off-target sites are near-matches elsewhere in the genome that the guide could also cut. Predict them by searching the genome for sequences that differ from the guide by a few mismatches next to a PAM, and score each candidate by how many close off-targets it has and where they fall (an off-target inside another gene is worse than one in intergenic space). Prefer guides with few off-targets and no perfect or single-mismatch off-targets in coding regions.

Step 5: Validate across more than one tool

On-target and off-target scores differ between algorithms, so validate the shortlist across independent tools rather than trusting one. Established options include CRISPOR, CHOPCHOP, and CRISPick, each of which combines specificity and efficiency scoring. A common pitfall is a guide that scores well but matches a pseudogene or a repeated element, which inflates apparent off-targets or causes cutting at unintended copies; cross-tool checks and inspecting where the off-targets land catch this.

Modality notes

The design rules shift with the CRISPR modality:

• Knockout: target an early constitutive exon; a frameshift near the N-terminus is most disruptive
• Knock-in / HDR: place the cut within a few bases of the edit, and provide a repair template with homology arms
• Base editing: the target base must fall inside the editor’s activity window (a few bases within the protospacer), and nearby bystander bases can be co-edited
• Prime editing: design the pegRNA (spacer plus a 3′ extension encoding the edit and a primer binding site); it does not depend on a nearby PAM the way HDR does
• CRISPRa / CRISPRi: target the promoter or transcription start site with catalytically dead Cas9 (dCas9); no cut is made

Common pitfalls

• No valid PAM at the desired cut position, so that site cannot be reached with the chosen enzyme
• A guide with a perfect or single-mismatch off-target inside another gene
• A guide that matches a pseudogene or repeat, inflating off-target counts
• For HDR, placing the cut too far from the intended edit
• A poly-T stretch in the spacer that terminates guide transcription

A worked example

Suppose you want a clean knockout of a gene. You pull the gene from a reference database, select an early constitutive exon, scan both strands for 20-nucleotide spacers next to an NGG PAM, score the candidates with Rule Set 2, filter out any guide with a perfect or single-mismatch off-target in a coding region, and confirm the top two or three guides in CRISPOR and CHOPCHOP before ordering. The result is a small, ranked set of guides with documented on-target and off-target scores.