Design a CRISPR knockout

A knockout means cutting, letting the cell repair badly, and ending up with a frameshift that destroys the reading frame. The design job is picking the guide most likely to cut where you want and least likely to cut anywhere else.

Pick a target

Three routes in, all equivalent downstream.

By symbol. Type the gene and press Fetch. Human and mouse resolve through Ensembl; other organisms go via UniProt and EMBL.

TP53

By accession. ENST…, ENSG…, NM_…, NR_…, XM_…, XR_….

By sequence. Paste raw DNA or FASTA. Headers and whitespace are stripped and anything outside ACGTN is dropped, so pasting straight out of a record works.

The cap is 1 Mbp. Human genes reach ~2.4 Mbp, so for the largest ones paste the exon or region you are editing.

Choose a nuclease

Default is SpCas9, and it is the right default. Reach for Cas12a when the region is AT-rich enough that NGG sites are scarce, when you want staggered ends, or when you are multiplexing — Cas12a processes its own array.

Base-edit mode is SpCas9 only. See base editing.

Screen against a genome

This is the step most worth not skipping. Leave Organism unset and the engine only checks for off-targets inside the sequence you pasted. Set it and your top-ranked guides are screened against a real reference.

Two mechanics worth knowing. Only the top ten ranked guides are screened, because each genome query costs a few seconds. And the index is built on first use — about one second per megabase, so seconds for E. coli and around three minutes for human. Until it finishes, genome_index_status reads building and the genome columns are blank. That is not a failure; run it again shortly.

Add a gene symbol as well to get exon context on each guide, resolved from Ensembl — useful for preferring an early constitutive exon, where a frameshift destroys the most protein.

Read the table

The table is twenty columns wide and scrolls sideways. Seven of them decide a choice:

A good knockout guide is high composite, unique in both off-target columns, high FS %, and in an early constitutive exon.

The remaining thirteen carry the identity of each guide (rank, strand, position, spacer, PAM, GC%), the predicted repair outcome (top indel), base-editor windows, exon context, the two cloning oligos, per-row vendor buttons, and any flags.

Two checkboxes above the table cut it down: hide flagged (GC / poly-T) drops guides with composition problems, and unique only (no off-target) keeps only guides with no hit anywhere they were screened.

Copy the methods paragraph

Above the table, Materials & methods (copy-paste) generates a written methods paragraph from the run you just did — enzyme, scoring provenance with citation, the selected guide, the cloning vector, and the honest limitation. One button copies it. From the EGFP run:

Guide RNAs (n=50) were designed in silico for SpCas9 (NGG PAM). On-target
activity was scored from Doench-style sequence features, and off-target
potential within the provided sequence by the cutting-frequency-determination
(CFD) matrix (Doench et al., Nat. Biotechnol. 2016); knockout likelihood was
estimated from the predicted indel spectrum (frameshift fraction and
out-of-frame dominance). The top-ranked guide (GAAGGGCATCGACTTCAAGG, sense
strand, composite 0.86) was selected, and cloning oligos were generated for the
standard BbsI/BsmBI vector. Off-target assessment is limited to the input
sequence and is not genome-wide. Analyses were performed with TuringDNA
(turingdna.com).

It states its own limits — note it says off-target assessment is not genome-wide, which is true for a run with no organism set.

Order the oligos

Pick your cloning vector and the oligo columns re-render with the right overhangs:

Each row’s Order cell has three buttons — IDT, SYN, TWST — that hand that guide’s oligo pair to IDT, Synthego or Twist Bioscience. Above the table, Order all guides from does the same for the whole set. Export gives the table as Excel or CSV, with the research-use-only statement attached.

Check prior art

IP radar searches Europe PMC, and PatentsView where a key is configured, for published work on the gene, the point mutation, or the 20-mer spacer itself. It is a literature and patent search, not legal advice — use it to find out whether a guide is already characterised, which is often more useful than the patent question.

Log what happened

After the bench, Log editing efficiency records the measured result against the design. This is the only step that produces information the scoring did not already have, and it is what makes the next round’s ranking better than this one’s. Everything upstream is prediction.