---
title: CRISPR on the platform
description: >-
  Guide RNA design, off-target analysis, and CRISPR screen hit identification
  for genome editing experiments.
tags:
  - crispr
---

Design and analyze genome editing experiments — score guide RNAs for a target gene, check candidate guides for off-target risk against the reference genome, and call hits from pooled CRISPR screens. Reach for these tools when you are doing functional target validation or knockout experiment design, upstream of picking a guide for the bench. CRISPR sits in the Target Identification stage of discovery, between gene-level genomics and experimental validation. See the [Targets & Modalities overview](/docs/targets-modalities) for where this fits.

> **What these three tools actually do, vs. what their names imply**

> All three CRISPR tools are simpler than their descriptions suggest. `crispr_design_guides` does not resolve a gene symbol to a sequence — it requires `target_sequence` and returns zero guides with an error if that's missing, even when `target_gene` is set; look up the sequence first (for example with a genomics lookup tool). On-target scoring is a GC/homopolymer/seed-region heuristic, not a trained model. `crispr_offtarget` does not search the reference genome — it extracts one guide from your `target_sequence` and scans that same short sequence for near-matches, so it can only find off-targets that happen to fall inside the sequence you submitted. `crispr_screen_analyze` does not run MAGeCK or BAGEL2 — it computes a median log2 fold-change per gene with a Benjamini-Hochberg FDR correction, and it infers control vs. treatment positionally (first numeric count column is control, the rest are treatment) rather than by name. The `organism` parameter has no effect on any of the three; genome build is a label on the output, not a species selector.

## CRISPR overview

The CRISPR toolkit provides three complementary tools for genome editing experiment design and analysis. Guide design and off-target prediction use direct computation, returning results in seconds. Screen analysis uses Modal GPU compute for statistical hit calling on large-scale datasets.

## Architecture

```
User Query (gene symbol, guide sequence, screen data)
  --> Intent Detection
       |- Guide design --> crispr_design_guides (5-15s)
       |- Off-target check --> crispr_offtarget (5-15s)
       +- Screen analysis --> crispr_screen_analyze (30-90s, Modal)
  --> Structured Results with Scores and Rankings
```

## Tool inventory

| Tool | Name | Description | Compute |
|------|------|-------------|---------|
| Guide RNA Design | `crispr_design_guides` | Designs guide RNAs for a target gene or sequence with on-target efficiency scoring, GC content, and PAM site identification | Direct API |
| Off-Target Analysis | `crispr_offtarget` | Scans a submitted target sequence (not the full reference genome) for near-matches to an extracted guide, with mismatch counts and a CFD-style score | Direct API |
| Screen Analysis | `crispr_screen_analyze` | Runs a median-log2FC / Benjamini-Hochberg heuristic (not MAGeCK itself) on pooled CRISPR screen read count data to rank gene hits | Modal GPU |

## Capabilities

### Guide RNA design

Given a target sequence, `crispr_design_guides` identifies all valid PAM sites on both strands, extracts 20-nt guide sequences, and scores each with a sequence-composition heuristic (GC-content band, homopolymer-run penalty, seed-region G/T bias) rather than a trained on-target model. Results include guide sequence, PAM, genomic position, strand, GC content, and on-target score. Guides are ranked by that heuristic score.

### Off-target prediction

`crispr_offtarget` takes a `target_sequence`, extracts the first valid guide it finds in it, and scans that same sequence on both strands for near-matches with up to 4 mismatches, scoring each with a simplified CFD (Cutting Frequency Determination)-style model that penalizes seed-region mismatches more heavily. It reports mismatch positions, the off-target sequence, position, strand, and a CFD score.

### CRISPR screen analysis

`crispr_screen_analyze` accepts a read-count table from a pooled CRISPR library screen (uploaded to S3), computes a per-guide log2 fold change, aggregates it to a per-gene median with a z-test, and applies Benjamini-Hochberg FDR correction — a method the backend describes as "inspired by MAGeCK," not MAGeCK's actual RRA algorithm. It treats the first numeric count column as control and the remaining numeric columns as treatment.

## When to use each tool

| I want to... | Use this tool |
|--------------|---------------|
| Design guides for a gene knockout | `crispr_design_guides` |
| Check if a guide has off-target risk | `crispr_offtarget` |
| Find the best guide with fewest off-targets | `crispr_design_guides` then `crispr_offtarget` |
| Identify essential genes from a screen | `crispr_screen_analyze` |
| Validate screen hits | `crispr_screen_analyze` then genomics tools |

## When to use other tools instead

| I want to... | Use this instead |
|--------------|------------------|
| Look up gene coordinates | An Ensembl or NCBI Gene lookup ([Genomics](/docs/genomics)) |
| Find protein structure for a target | [AlphaFold DB](/docs/alphafold) or [RCSB PDB](/docs/pdb) |
| Assess target druggability | OpenTargets associations ([Genomics](/docs/genomics)) or a Target Passport |
| Search literature on a gene | [PubMed](/docs/pubmed) |
| Predict protein function | [UniProt](/docs/uniprot) |

## Pipeline position

The CRISPR toolkit sits in the **Target Identification** stage of the drug discovery pipeline:

```
Target Identification
  |- Genomics (gene information)
  |- CRISPR (functional validation) <--
  |- Epigenomics (regulatory context)
  +- Literature (evidence gathering)
```

Guide design is typically preceded by target gene identification (via genomics or literature tools) and followed by experimental validation. Screen analysis results feed back into target prioritization.

## Citations

> Doench JG, et al. Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9. *Nat Biotechnol*. 2016;34(2):184-191. doi: [10.1038/nbt.3437](https://doi.org/10.1038/nbt.3437)

> Li W, et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. *Genome Biol*. 2014;15(12):554. doi: [10.1186/s13059-014-0554-4](https://doi.org/10.1186/s13059-014-0554-4)

> Hsu PD, et al. DNA targeting specificity of RNA-guided Cas9 nucleases. *Nat Biotechnol*. 2013;31(9):827-832. doi: [10.1038/nbt.2647](https://doi.org/10.1038/nbt.2647)

## Chat usage

### Example prompts

- "Design CRISPR guides for KRAS exon 2"
- "Check off-target sites for GACTTCGATCGATCGATCGA with 4 mismatches allowed"
- "Analyze my CRISPR screen to find genes essential for cell survival"
- "Design SaCas9 guides targeting MYC"
- "What are the top guides for knocking out CDK4 with minimal off-target risk?"

### Workflow chaining

- **Feeds into**: Genomics (gene validation), Target Passport (druggability), Literature (essentiality evidence), Protein structure tools (target characterization)
- **Accepts from**: Genomics (gene symbols, coordinates), Literature (candidate gene lists), Epigenomics (chromatin state for guide accessibility), OpenTargets (target prioritization)

## FAQ

### Which CRISPR systems does the toolkit support?

  `crispr_design_guides` and `crispr_offtarget` accept `cas_type` as `cas9`, `cas12a`, or `cas13`, but the backend's PAM lookup table is keyed on `SpCas9`/`SaCas9`/`Cas12a`/`AsCas12a`/`CjCas9` — none of which match the lowercase values the tools actually send, so PAM selection always falls back to NGG (SpCas9) regardless of which `cas_type` you choose. Provide a custom `pam` string if you need a non-NGG PAM.

### How is guide RNA on-target activity scored?

  On-target scoring is a sequence-composition heuristic (GC-content band, homopolymer-run penalty, seed-region G/T bias), not the Rule Set 2 (Doench 2016) model. Scores range from 0 to 1, with higher values indicating a more favorable sequence composition by this heuristic — not a benchmarked cutting-efficiency prediction.

### What is the off-target analysis based on?

  `crispr_offtarget` does not search the reference genome. It extracts one guide from the `target_sequence` you submit and scans that same sequence (both strands) for near-matches with up to 4 mismatches, scoring each with a simplified CFD-style model. Results include mismatch count, position, strand, and CFD score — not genomic coordinates or nearest-gene annotation.

### What is MAGeCK, and does this tool run it?

  MAGeCK (Model-based Analysis of Genome-wide CRISPR-Knock-out) is a published statistical algorithm for identifying essential genes from CRISPR screen data. `crispr_screen_analyze` does not run it — it computes a per-guide log2 fold change, a per-gene median with a z-test, and Benjamini-Hochberg FDR correction, a method the backend itself describes as "inspired by MAGeCK." The `method` parameter (`mageck`/`bagel2`) has no effect on which method actually runs.

## Read next