bio-crispr-screens-prime-editing-screens

What this skill does

## Version Compatibility

Reference examples tested with: PRIDICT2 v1.0+ (https://github.com/uzh-dqbm-cmi/PRIDICT2), CRISPResso2 2.2.14+, pandas 2.2+, biopython 1.83+, numpy 1.26+.

Before using code patterns, verify installed versions match. If versions differ:
- CLI: `python pridict2_pegRNA_design.py single --help`; `python pridict2_pegRNA_design.py batch --help`
- Web: PRIDICT2 web interface at https://pridict.it/

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

## Prime-Editing Screen Analysis

**"Design or analyze a pooled prime-editor screen"** -> Design pegRNAs (spacer + scaffold + PBS + RTT) for intended edits, predict efficiency with PRIDICT2, filter pre-synthesis to efficient candidates, install variants in the screen, quantify intended-edit vs scaffold-incorporation vs indel via CRISPResso2, and aggregate to per-variant fitness scores.

- Python: `PRIDICT2` for pegRNA efficiency prediction
- CLI: `CRISPResso --prime_editing_pegRNA_*` for amplicon-level analysis
- Workflow: pegRNA library design -> PRIDICT2 filtering -> screen execution -> CRISPResso2 quantification -> per-variant scoring

## Prime Editor Chemistry Comparison

| Editor | Year | Mechanism | Indel rate | Use when |
|--------|------|-----------|------------|----------|
| PE2 (Anzalone 2019) | 2019 | nCas9-RT fusion + pegRNA | 1-3% | Standard PE; lowest indel rate |
| PE3 | 2019 | PE2 + nick of opposite strand by additional sgRNA | 2-5% | Higher editing efficiency, slightly more indels |
| PE3b | 2019 | PE3 with edit-blocking ssgRNA | 1-3% | When PE3's added nick risks unwanted indels |
| PEmax (Chen 2021) | 2021 | Engineered RT + nCas9 | 1-2% | Higher editing rate per pegRNA |
| PEAR (Erwood 2023) | 2023 | PE with optimal pegRNA scaffold | 1-2% | Improved PE scaffold |
| PE5max (Chen 2021) | 2021 | PEmax with engineered scaffold variants | 1% | Highest efficiency at favorable sites |
| Dual-pegRNA / PE6 (2024) | 2024 | Twin pegRNA system | Variable | Specific applications |

**Decision rule:** For pooled screens at scale, PE2 or PEmax (less RAM-intensive in cells) is preferred over PE3 (additional sgRNA complicates library architecture). For specific high-efficiency edits, PEmax + PRIDICT2-optimized pegRNA.

## pegRNA Architecture

A pegRNA contains four critical elements that determine efficiency:

```
5'  SPACER (20 nt)  -- standard sgRNA spacer; defines target locus via NGG PAM
    +
    SCAFFOLD (~80 nt) -- canonical or engineered scaffold (Chen 2021 has improved scaffold)
    +
    PBS (Primer Binding Site, 8-15 nt) -- complements protospacer downstream of cut site
    +
    RTT (Reverse Transcription Template, 10-30 nt) -- encodes intended edit; copied by RT
3'
```

**Key design parameters:**
- **PBS length:** 11-13 nt typical; longer for high-GC contexts; PBS GC fraction critical (35-65% target)
- **RTT length:** 10-20 nt typical; longer for distant edits (10+ bp away from cut)
- **RTT-edit position:** intended edit at position 4-30 from cut site
- **Scaffold:** standard sgRNA scaffold OR Chen 2021 engineered scaffold (5-10% higher editing)

## PRIDICT and PRIDICT2 pegRNA Efficiency Prediction

**Mathis N et al 2023 *Nat Biotechnol* 41:1151 (PRIDICT v1) / 2025 *Nat Biotechnol* 43(5):712 (PRIDICT2; published online June 2024)** developed deep-learning predictors of per-pegRNA editing efficiency. PRIDICT2 is the current state of the art.

```bash
# PRIDICT2 is invoked via CLI: pridict2_pegRNA_design.py
# Single sequence input:
python pridict2_pegRNA_design.py single \
    --sequence-name BRCA1_c5135 \
    --sequence "AGCAGCCT(C/T)CTGAATGCCC...60nt_context" \    # parens = intended edit
    --output-dir predictions/ \
    --use_5folds                                              # 5-fold ensemble averaging

# Batch input from CSV:
python pridict2_pegRNA_design.py batch \
    --input-fname variants_to_design.csv \                    # CSV: sequence_name, sequence
    --output-dir predictions/ \
    --cores 4 \
    --summarize                                               # generate summary table

# Output: per-pegRNA predictions in predictions/<sequence_name>/
# Columns: PBS_sequence, PBS_length, RTT_sequence, RTT_length, predicted_editing_efficiency,
#          predicted_indel_rate, deep_ensemble_score, etc.
```

**Loading PRIDICT2 results in Python:**

```python
import pandas as pd
from pathlib import Path

def load_pridict2_predictions(prediction_dir):
    '''Load PRIDICT2 batch outputs from prediction_dir/'''
    summary = pd.read_csv(Path(prediction_dir) / 'pridict2_summary.csv')
    # summary has columns: sequence_name, PBS, RTT, predicted_efficiency, predicted_indel, etc.
    return summary
```

**Key determinants of PE efficiency (Mathis 2024 PRIDICT2):**

| Feature | Effect on efficiency |
|---------|----------------------|
| PBS GC content | 40-55% optimal; high GC slows annealing |
| PBS length | 11-13 nt optimal; longer for high-GC PBS |
| RTT length | 10-20 nt typical; trade-off between coverage and processivity |
| Edit position in RTT | Closest to PBS = highest efficiency |
| Chromatin context | Open chromatin = 2-5x higher efficiency than closed |
| Cell line / Cas9 expression | Variable; piloting required |
| Cell cycle phase | S/G2 = higher efficiency |

**Critical insight from Mathis 2024:** Chromatin context is the dominant determinant. Sequence-based predictions like PRIDICT under-predict at silenced loci and over-predict at open chromatin. For genome-scale screens, validate predictions empirically at representative loci.

## PRIME Pooled Screen Methodology

**Erwood S, Doman JL et al 2023 *Nat Biotechnol* 41:885** established the PRIME pooled-screen methodology (earlier 2022 bioRxiv preprint):

- pegRNA library covering thousands of intended variants
- Filter pegRNAs to PRIDICT2 efficiency >50% (or pilot top 25%)
- Lentiviral delivery at standard MOI 0.3 in PE-expressing cell line
- Selection on integration marker
- Time-course screen for variant function (e.g., drug sensitivity)
- Endpoint amplicon sequencing of each pegRNA target locus
- CRISPResso2 quantification of intended-edit %
- MAGeCK / drugZ-style hit calling on edit-efficient pegRNAs

**Quantified scale:** ~3,699 ClinVar variants installed in a single PRIME screen (Erwood/Doman 2023 *Nat Biotechnol* 41:885), with editing efficiency >5% at >50% of pegRNAs (validation cohort).

## MOSAIC In Situ Saturation Mutagenesis

**MOSAIC (Hsu JY, Lam KC, Shih J, Pinello L, Joung JK 2024 bioRxiv 10.1101/2024.04.25.591078)** is a higher-throughput variant of PRIME with multiplexed read-out:

- Tile pegRNAs across protein domains for systematic mutagenesis
- Saturation: every possible amino acid change in a region
- Identify drug-resistance variants in real-time
- Smaller per-variant cell numbers (more variants total)

**Use case:** Cancer-drug-resistance variant scanning; protein-domain function mapping.

## Run PRIDICT2 on a Custom pegRNA Library

**Goal:** Predict editing efficiency for thousands of pegRNAs before library synthesis.

**Approach:** Build a CSV with one row per intended edit (sequence + edit notation), run PRIDICT2 in batch mode, parse the per-pegRNA efficiency summary, and filter to candidates above the chosen efficiency threshold.

```bash
# Step 1: prepare batch input CSV (sequence_name, sequence with (REF/ALT) edit notation)
cat > variants.csv <<EOF
sequence_name,sequence
BRCA1_R71X,AGCAGCCT(C/T)CTGAATGCCC...
MLH1_c677,GAGCTGAGC(A/G)GAGGCTCTTGAAGC...
EOF

# Step 2: run PRIDICT2 batch
python pridict2_pegRNA_design.py batch \
    --input-fname variants.csv \
    --output-dir predictions/ \
    --cores 8 \
    --summarize
```

```python
# Step 3: parse and filter
import pandas as pd
predictions = pd.read_csv('predictions/pridict2_summary.csv')

# Filter to pegRNAs with predicted efficiency > 50% (Mathis 2024 threshold)
filtered = predictions[predictions['predicted_editing_ef