Base Editing Analysis - Usage Guide

Overview

Decision-grade analysis of base-editor variant-function screens. Covers base-editor library design, the Hanna 2021 BRCA1/2 SNV-scanning methodology (Cell 184:1064), Cuella-Martin 2021 complementary DDR-gene saturation screen (Cell 184:1081-1097), CBE vs ABE chemistry selection (BE3/BE4 vs ABE7.10/ABE8.20/ABE8e), editing window math (positions 4-8 from PAM-distal end), bystander attribution strategies, editing-efficiency filtering before hit calling, indel-byproduct interpretation, and the Broad be-validation-pipeline notebooks for CRISPResso2 post-processing.

Prerequisites

conda install -c bioconda crispresso2   # not on PyPI pandas biopython numpy scipy
# BE-Hive for editing efficiency prediction
git clone https://github.com/maxwshen/be_predict_bystander   # BE-Hive is not on PyPI
# Broad end-to-end pipeline
git clone https://github.com/broadinstitute/be-validation-pipeline
cd be-validation-pipeline && pip install -r requirements.txt

Required inputs: - Amplicon sequencing FASTQ (per sgRNA or per pool) - Library file: per-sgRNA spacer, target base, target amino acid, predicted bystander pattern - BE chemistry vector (CBE or ABE) - (For variant attribution) ClinVar / COSMIC variant annotation

Quick Start

Tell the AI agent what to do: - "Design a CBE library tiling BRCA1 exons 1-23 with 10-15 sgRNAs per amino acid; flag intended target Cs and bystander Cs in the editing window" - "Run editing-efficiency filtering on pilot library counts; drop sgRNAs <30% target editing" - "Score variant function per amino acid; deconvolute bystander confounding via allele-frequency tables" - "Reconcile a screen hit between Hanna's CBE methodology and orthogonal prime-editor follow-up" - "Choose BE vs PE for installing a C->T variant: BE3 if target at pos 5 with no bystanders; PE2 if zero-bystander requirement" - "Diagnose: my BE sample has 60% editing but 35% are indels -- is this Cas9 contamination?"

Example Prompts

Library Design

"Design a CBE saturation library tiling BRCA1 RING domain (amino acids 1-100). 10-15 sgRNAs per amino acid where at least one C in the editing window (positions 4-8) hits the target codon. Annotate each sgRNA with predicted target + bystander variants. Output library.tsv with sgRNA, target_aa, target_variant, bystander_variants columns."

"I need to install MLH1 c.677A>G as a single intended variant. CBE won't work (need ABE). Find ABE7.10 or ABE8e sgRNAs that place A at position 5 with no bystanders. If no zero-bystander spacer exists, list candidates sorted by bystander_count and recommend prime editor as alternative."

Editing Efficiency Filtering

"Run CRISPResso2 on my pilot timepoint samples. Compute target editing % per sgRNA. Keep sgRNAs >30% target editing for the primary screen; flag >50% as validation-grade. Output filtered library and editing efficiency report."

"My library shows median 22% editing across guides. Diagnose: cell-line BE activity issue, vector mismatch, or BE chemistry choice (CBE vs ABE)?"

Hit Calling for Variant Function

"Apply MAGeCK MLE to the BE screen counts (vehicle vs PARPi drug). Use only editing-efficient sgRNAs. Aggregate per-sgRNA LFC to per-variant scores; deconvolute bystanders via allele tables; output per-variant fitness with confidence based on the number of sgRNAs hitting each variant."

"Compare drugZ vs MAGeCK MLE on the same BE drug-modifier screen; identify resistance and sensitivity variants using the Hanna 2021 CBE variant-scanning approach."

Bystander Deconvolution

"From CRISPResso2 allele tables, separate reads by edit pattern: target only, target+bystander_1, target+bystander_2. Compute per-pattern fitness contribution; flag variants where bystander dominates."

"For BRCA1 R71 -> R71X variant, I have 5 sgRNAs with different bystander patterns. Identify variants attributable to R71X alone by finding consistent signal across diverse bystander backgrounds."

Method Comparison

"Compare CBE BE3 vs BE4max vs eA3A-BE3 vs evoCDA-BE for tiling BRCA1. Score: editing window width, bystander rate, indel byproduct, target-base preference (TC vs other)."

"Cross-validate Hanna 2021 BRCA1 BE screen hits with parallel PE screen at the same variants. Report concordant variants and BE-only (likely bystander-confounded) hits."

Diagnostics

"My BE sample has 70% editing but 25% indels. Compute substitution-vs-indel ratio; if <3, diagnose as Cas9 contamination."

"Allele table shows 35% target+bystander double-edit. Deconvolute: how much is target-attributable vs bystander-driven?"

What the Agent Will Do

Design BE library: for each target amino acid, find NGG-PAM spacers with target base at editing window positions 4-8; minimize bystanders
Annotate each sgRNA: target_variant, bystander_pattern (other edits in window)
Order library; receive plasmid pool; sequence to verify Gini <0.1
Lentiviral package and infect at MOI 0.3 in BE-validated cell line (HEK293T, U2OS, K562)
Pilot timepoint amplicon sequencing of ~20 representative loci
Run CRISPResso2 with --base_editor_output, --conversion_nuc_from C --conversion_nuc_to T (or A->G for ABE)
Compute editing efficiency per sgRNA; filter to >30% (primary) or >50% (validation)
Run main screen with vehicle vs drug treatment
Per-sgRNA hit calling via MAGeCK MLE or drugZ
Aggregate to per-variant scores; deconvolute bystander via allele tables
Cross-check substitution-vs-indel ratio per sgRNA (>10 for clean BE)
Validate top hits via orthogonal prime-editor or arrayed BE
Annotate against ClinVar / COSMIC for clinical interpretation

Tips

Editing efficiency filtering is non-negotiable. Unedited reads carry no biological perturbation; including <30% efficiency sgRNAs adds noise and dilutes signal. A 30% primary / 50% validation split is the common working convention.
Bystander confounding is the central interpretation challenge in BE screens. Plan the library with multiple sgRNAs per target (10-15 in saturation designs) so bystander patterns vary; consistent signal across diverse bystanders attributes to target.
For pristine variant-function calls (clinical-grade), validate with prime editor (zero bystanders). Convergent BE + PE signal is the gold standard for variant pathogenicity in pooled screens.
Cell-line BE activity varies; pilot in your target line before designing the full library. Median library editing <30% is a cell-line issue, not a library issue.
ABE7.10 has lower indel rate (<2%) than CBE (5-10%); when target is A->G, ABE is the cleaner choice.
For C->T variants with target at position 4-5 (PAM-distal end) and bystander at position 6-7, both will edit; this is the most common interpretation trap.
Editor chemistries have different windows: BE3/BE4 = 4-8, eA3A = 5-7 (narrower; good for bystander minimization), ABE7.10 = 4-7 (narrowest ABE), SpABE8e = 4-8 (Richter 2020; more processive, so more bystander editing within the window), evoCDA = 1-9 (broadest; high bystander rate).
Run substitution-vs-indel ratio per sgRNA as a per-sample QC. <3 means Cas9-like activity (vector mismatch or contamination); >10 means clean BE.
The Broad be-validation-pipeline notebooks are the canonical post-processing reference for BE amplicon data; reuse them before writing custom parsers. They do not do hit calling, so score the screen with drugZ or MAGeCK.
For drug-modifier BE screens (like Hanna 2021 PARPi), drugZ is more sensitive than MAGeCK for chemogenomic interactions.

Chemistry Cheat Sheet

Need	Editor	Why
C->T at TC context	BE3, BE4max, eA3A-BE3	eA3A-BE3 narrowest window
C->T at any context	BE4max	Standard CBE
A->G	ABE7.10, ABE8.20, ABE8e	ABE8e fastest editing
Multi-base / no bystander	Prime editor	See [[prime-editing-screens]]
Transversions (C->G, C->A)	CGBE1, GBE	Rare use; less mature
Saturation mutagenesis	BE3 + PE for orthogonal validation	Combined coverage

Validation Strategy

Tier	Validation requirement
Tier 1 (high confidence)	BE + PE concordant at same variant + arrayed confirmation
Tier 2 (medium)	BE alone, multiple sgRNAs converge despite bystander differences
Tier 3 (exploratory)	Single sgRNA hit; bystander confounded; not interpretable

crispr-screens/crispresso-editing - CRISPResso2 modes and allele tables
crispr-screens/library-design - base-editor library design
crispr-screens/prime-editing-screens - Orthogonal PE for variant attribution
crispr-screens/hit-calling - Variant-level hit aggregation
crispr-screens/screen-qc - Editing-efficiency QC
crispr-screens/drugz-chemogenomic - drugZ for BE drug-modifier screens
crispr-screens/mageck-analysis - MAGeCK MLE for BE screen sgRNA-level analysis
clinical-databases/clinvar-lookup - Pathogenicity annotation
variant-calling/variant-annotation - VEP for predicted protein effects