Bio Workflows Crispr Editing Pipeline

Version Compatibility

Reference examples tested with: BioPython 1.83+, pandas 2.2+, matplotlib 3.8+.

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures
• CLI: <tool> --version then <tool> --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

This workflow coordinates the five genome-engineering skills; it does not re-implement their scoring. Real on-target ranking comes from CRISPOR (context-valid model), off-target nomination from Cas-OFFinder/CRISPRme, base-editor outcomes from BE-Hive, and prime-editing ranking from PRIDICT/DeepPrime -- the embedded code is illustrative orchestration only.

CRISPR Editing Pipeline

"Design a complete CRISPR editing experiment for my target" -> Run guide design -> off-target assessment -> edit-modality selection -> template/donor design -> validation, applying a QC checkpoint at each handoff and routing every mechanic to the relevant genome-engineering skill.

• Python: orchestrate the stages; enumerate/filter candidate guides with Bio.Seq
• CLI/web: CRISPOR (on-target + off-target), Cas-OFFinder/CRISPRme (off-target), BE-Hive, PRIDICT

The Single Most Important Modern Insight -- the pipeline is a chain of handoffs, each with a checkpoint, and the pivotal decision is the edit modality

A CRISPR experiment fails most often not at one step but at a handoff where an unstated assumption carries through: a guide picked by on-target score that turns out non-specific, an "efficient" guide that never knocks out the protein, a base edit reported by efficiency that is a genotype soup, an HDR donor with no blocking mutation whose edit is silently re-cut. The workflow's job is to make each handoff explicit and gated. The pivotal branch is which edit modality: a transition (C->T/A->G) is usually a base-editing job; any other small precise edit is prime editing; a knockout is a plain nuclease; a large or non-transition insertion is HDR (or PE+integrase). Choosing the modality first reframes every downstream step. The cross-cutting traps the checkpoints exist to catch: on-target activity != specificity (two separate axes), efficient editing != knockout (frameshift fraction and NMD-competent exon biology decide it), base-editor efficiency != purity (bystanders), a donor without a blocking mutation self-destructs (re-cutting reads out as failed HDR), and predicted != detected != validated for off-targets.

Edit-Modality Decision Tree (the pivotal branch)

Workflow Overview

Target gene / position
        |
        v
[1. Guide design] ----> CRISPOR (context-valid on-target) + outcome model (Bae microhomology / inDelphi)
        |                CHECKPOINT: shortlist 3-6, frameshift-rich, early constitutive exon
        v
[2. Off-target assessment] ----> Cas-OFFinder (+bulges) / CRISPRme (variant-aware) + CFD
        |                CHECKPOINT: no low-mm high-CFD off-target in a gene; predicted->detected->validated
        v
    DECISION: which edit modality?
        |
    +----------+-------------+--------------+-------------+
    v          v             v              v             v
[3a. KO]   [3b. Base edit] [3c. Prime edit] [3d. HDR knock-in]
 frameshift  window+purity   pegRNA panel     donor + codon-checked block
        |          |              |              |
        v          v              v              v
[4. Validation] ----> amplicon deep-seq (CRISPResso2); report purity/indels; state LoD

Stage 1 -- Guide Design (-> grna-design)

Goal: A shortlist of 3-6 specificity-checkable guides whose predicted repair outcome is frameshift-rich, in an early constitutive NMD-competent exon.

Approach: Establish the delivery context (it sets the valid on-target model and the hard filters), enumerate PAMs on both strands, drop TTTT/GC-extreme guides, rank on-target with the context-valid model via CRISPOR (not a hand-rolled score), and rank knockout candidates by predicted frameshift/out-of-frame fraction (Bae microhomology / inDelphi). Checkpoint: carry 3-6 guides; do not commit on raw activity alone.

Stage 2 -- Off-Target Assessment (-> off-target-prediction)

Goal: Reject promiscuous guides and, for therapeutics, establish an evidence-laddered specificity profile.

Approach: Enumerate candidates with Cas-OFFinder including bulges and a relaxed PAM; rank by CFD; for a research knockout this in-silico pass is sufficient. For a therapeutic, run variant-aware nomination (CRISPRme vs gnomAD + individual), choose a high-fidelity nuclease in the delivery format used, and plan empirical discovery + amplicon validation. Checkpoint: on-target score does not predict specificity; treat predicted/detected/validated distinctly.

Stage 3 -- Modality-Specific Design

Goal: Produce the construct(s) for the chosen modality.

Approach: Branch by the decision tree. Knockout -> the frameshift-ranked guide. Base editing -> position the target base at the window peak, minimize bystanders, choose the editor variant, report the genotype spectrum (-> base-editing-design). Prime editing -> a PBS x RTT panel with PAM-disrupting/MMR-evading silent edits and a 3' motif, ranked by PRIDICT/DeepPrime (-> prime-editing-design). HDR -> the donor format for the cell type with a mandatory codon-checked blocking mutation and the cut within ~10 bp of the edit (-> hdr-template-design). Checkpoint: blocking mutation present and codon-checked; base-editing purity reported.