CRISPR FAQ
Frequently asked questions about CRISPR guide design, off-target analysis, and screen analysis tools on the platform.
CRISPR FAQ
These answers cover CRISPR guide design — supported Cas nucleases, on-target scoring reliability, GC content, and non-human organisms — along with off-target mismatch settings and MAGeCK screen analysis, including replicate requirements and CRISPRi/CRISPRa support.
What Cas nucleases are supported for guide design?
The guide design tool supports any Cas nuclease by specifying its PAM sequence. Built-in presets include SpCas9 (NGG), SaCas9 (NNGRRT), and AsCas12a/Cpf1 (TTTV). You can also provide custom PAM sequences using IUPAC nucleotide codes. The on-target scoring model is optimized for SpCas9 but provides reasonable predictions for other nucleases.
How reliable is the on-target score?
The on-target score is a sequence-composition heuristic (GC-content band, homopolymer-run penalty, seed-region G/T bias) rather than the trained Rule Set 2 model (Doench et al., 2016) — it is not benchmarked against experimental cutting-efficiency data, so no correlation figure applies to it. Treat the ranking as a rough sequence-quality filter, not a validated activity prediction, and always confirm with experimental validation for critical experiments. Factors like chromatin accessibility, which sequence-based heuristics can't capture either way, also affect cutting efficiency in cells.
How many mismatches should I allow in off-target analysis?
The default of 3 mismatches provides a good balance between comprehensiveness and specificity. At 3 mismatches, Cas9 cutting is generally negligible, but reporting these sites helps identify potential risks. Increasing to 4 or 5 mismatches dramatically increases the number of reported sites (often thousands) and is usually unnecessary for standard experiments. For therapeutic applications where safety is critical, consider allowing 4 mismatches.
Can I design guides for non-human organisms?
Yes, in the sense that guide design and off-target scanning both operate directly on whatever target_sequence you submit — there's no species-specific reference lookup involved, so any organism's sequence works. The organism parameter itself has no effect on the result; it isn't read by the backend. Since scoring is a generic sequence-composition heuristic rather than a species-trained model, there's no human-specific bias to correct for.
What is the optimal GC content for guide RNAs?
Guides with 40-60% GC content generally perform best. Very low GC content (<30%) can reduce guide stability and binding efficiency, while very high GC content (>70%) can cause secondary structures that interfere with Cas9 loading. The guide design tool reports GC content for each candidate and can filter by minimum and maximum GC thresholds.
Does this tool run MAGeCK?
No. crispr_screen_analyze runs a simpler method the backend describes as "inspired by MAGeCK": per-guide log2 fold change, aggregated to a per-gene median with a z-test, and Benjamini-Hochberg FDR correction. It does not run MAGeCK's Robust Rank Aggregation (RRA) algorithm, and the method parameter (mageck/bagel2) has no effect on which method runs. If you need MAGeCK's actual RRA statistics, run MAGeCK yourself on the exported count table.
What is the minimum number of replicates needed for screen analysis?
MAGeCK can run with a single replicate per condition, but results are more reliable with 2-3 biological replicates. With fewer replicates, the tool relies more heavily on the median ratio normalization and may have higher false positive rates. For publication-quality results, 3 replicates per condition is recommended.
Can I analyze CRISPRi or CRISPRa screens?
Yes. The screen analysis tool works with any pooled CRISPR library screen where read counts are measured before and after selection. CRISPRi (interference/knockdown) and CRISPRa (activation/overexpression) screens produce the same type of count data as knockout screens. The statistical analysis is identical -- only the biological interpretation of hits differs.
How do I select guides with minimal off-target risk?
The recommended workflow is:
- Design guides with
crispr_design_guides(aim for on-target score > 0.5) - Run
crispr_offtargeton the top candidates - Prefer guides with zero 1-mismatch off-target sites and fewer than 5 coding-region sites at 2 mismatches
- Among guides meeting these criteria, choose the one with the highest on-target score
Read next
Was this page helpful?