CRISPR Screen Pipeline - Usage Guide

Overview

End-to-end pipeline for pooled and single-cell CRISPR screens. Takes raw FASTQ files (or pre-counted matrices) through library QC, guide counting, six-stage quality control, optional copy-number correction for cancer lines, optional batch correction for multi-batch screens, design-matched hit calling across MAGeCK RRA/MLE, BAGEL2, drugZ, JACKS, and Chronos, and tier-based consensus. Branches into specialized workflows for single-cell Perturb-seq, combinatorial paralog screens, base-editor and prime-editor variant-function screens, and in vivo bottleneck-aware screens.

Prerequisites

# Core hit-calling tools (MAGeCK/mageck-vispr are on bioconda, NOT PyPI)
conda install -c bioconda mageck mageck-vispr
# BAGEL2 and JACKS are GitHub clones (the PyPI `jacks` is an unrelated package)
git clone https://github.com/hart-lab/bagel
git clone https://github.com/felicityallen/JACKS
pip install pertpy scanpy anndata
# drugZ (clone from GitHub)
git clone https://github.com/hart-lab/drugz
# Editing analysis (CRISPResso2 is on bioconda, not PyPI)
conda install -c bioconda crispresso2
# Chronos for DepMap-style cancer-line CN+quality jointly (GitHub, not PyPI)
pip install git+https://github.com/broadinstitute/chronos
# R packages
R -e "BiocManager::install('sva')"
# MAGeCKFlute was removed from Bioconductor at the 3.22 release
R -e "remotes::install_github('WubingZhang/MAGeCKFlute')"
# CRISPRcleanR and sceptre are GitHub packages, not Bioconductor/CRAN
R -e "devtools::install_github(c('francescojm/CRISPRcleanR', 'katsevich-lab/sceptre'))"
# Helpers
pip install pandas numpy scipy matplotlib seaborn scikit-learn biopython statsmodels

Required inputs: - FASTQ files (one per sample) OR pre-counted matrix - sgRNA library CSV with columns sgRNA, Gene, Sequence - Sample-to-condition mapping - For cancer-line screens: copy-number profile (WGS / SNP-array / ASCAT) - For multi-batch: batch annotation per sample - For in vivo: per-animal sample annotation

Quick Start

Tell the AI agent what to run: - "Run the end-to-end CRISPR screen pipeline on my Brunello dropout screen with vehicle and drug arms" - "Apply CRISPRcleanR copy-number correction before MAGeCK on my HER2+ SK-BR-3 panel" - "Pick MAGeCK RRA vs MLE vs BAGEL2 vs drugZ vs JACKS vs Chronos for my screen design and execute" - "Build tier-based consensus across MAGeCK + BAGEL2 + drugZ for high-stakes drug-target nomination" - "Branch the pipeline into perturb-seq-analysis for my single-cell CROP-seq dataset" - "Diagnose CEGv2 PR-AUC under 0.5 in my screen and recommend remediation"

Example Prompts

Standard Pooled Screen

"I have FASTQ from a Brunello dropout screen: plasmid + Day 0 + 3 replicates of Day 14 vehicle + 3 replicates of Day 14 drug. Run guide counting, six-stage QC, then MAGeCK test (drug vs vehicle) and drugZ in parallel. Output tier-2 consensus hits at FDR <0.05."

"Pick the appropriate hit-calling method for my time-course screen across 5 cell lines. Run MAGeCK MLE with cell-line covariates."

Cancer Cell Line + CN Bias

"My screen is in SK-BR-3 (HER2+, high-level ERBB2 amplification). Run CRISPRcleanR pre-hoc, then MAGeCK on corrected counts. Verify ERBB2 is no longer top hit."

"DepMap-style screen across 12 cancer cell lines, longitudinal sampling, matched WGS CN profiles. Use Chronos as the integrated CN + screen-quality + gene-effect model."

Chemogenomic Drug-Modifier

"PARPi sensitivity screen: vehicle vs olaparib at 14 days. Run drugZ with -i -o -c -x flags; expect sensitizers in DDR pathway (BRCA1/2, RAD51) and resistance in PARP1 paradox."

Time Course / Multi-Condition

"Build a MAGeCK MLE design matrix for Day 0 / 7 / 14 / 21 timepoints. Run mle; output per-condition beta scores; identify genes with monotonic depletion trends."

Specialized Branches

"Branch to perturb-seq-analysis: my data is CROP-seq with 10X 3' direct capture. Run sgRNA assignment, Mixscape escaper filter, then per-perturbation DE."

"Branch to combinatorial-screens: Inzolia 4-guide-array Cas12a library targeting 400 paralog pairs. Compute GI scores and identify synthetic-lethal pairs at GI z <-2."

"Branch to base-editing-analysis for my CBE saturation screen of BRCA1 exons 1-23 (Hanna 2021 methodology). Filter to >30% editing-efficient sgRNAs; aggregate to per-variant fitness."

"Branch to in-vivo-screens: focused 2,500-gene library in B16-OVA syngeneic with 10 mice per condition. Per-animal MAGeCK + Stouffer Z meta-analysis."

Diagnostics

"Screen passes Gini and Pearson but CEGv2 PR-AUC is 0.45. Diagnose Cas9 selection failure vs early timepoint vs library positioning vs CN bias and recommend remediation."

"My MAGeCK RRA says everything is significant at FDR <0.01. Diagnose heavy selection breaking median normalization; recommend switching to --norm-method control with NTCs or BAGEL2."

What the Agent Will Do

Inspect library file and FASTQ filenames; verify naming consistency
Run mageck count with sample labels and --trim-5 adapter
Inspect countsummary.txt for mapping rate, Gini, % zero per sample
Run six-stage QC (per [[crispr-screens/screen-qc]]): plasmid Gini, Day-0 coverage, replicate Pearson/Spearman, sequencing depth, CEGv2 PR-AUC against Hart 2017
If cancer cell line: apply CRISPRcleanR or Chronos for CN bias (per [[crispr-screens/copy-number-correction]])
If multi-batch: add batch covariate to MAGeCK MLE design matrix (per [[crispr-screens/batch-correction]])
Pick hit-calling method by design (RRA for two-condition; MLE for multi-condition; BAGEL2 for essentiality; drugZ for chemogenomic; JACKS for multi-screen joint; Chronos for cancer panels)
Execute primary method and at least one orthogonal method
Build tier-based consensus: Tier 1 = 3-method agreement, Tier 2 = 2 of 3
Cross-reference cancer-line amplification database to flag CN-suspect hits
Run MAGeCKFlute or VISPR for visualization + KEGG/Reactome enrichment
For specialized branches, hand off to perturb-seq-analysis, combinatorial-screens, base-editing-analysis, prime-editing-screens, or in-vivo-screens
Output per-tier hit list, volcano + rank plots, QC report, recommended orthogonal validation strategy

Tips

The single most common failure is skipping plasmid pool sequencing. Always sequence the plasmid library before infection and use it as baseline; comparing only to Day 0 conflates cloning bottleneck with biology.
For drug screens, always use vehicle (DMSO or carrier) as control, NOT Day 0. The drug-vs-Day-0 comparison conflates drug effect with normal proliferation.
Cancer cell line screens ALWAYS require copy-number correction. The Aguirre 2016 / Munoz 2016 amplicon artifact is universal; ERBB2 in HER2+, MYC in MYC-amplified, FGFR1 in head-and-neck cases are textbook.
Choose hit calling by experimental design, not by familiarity. MAGeCK RRA is great for two-condition essentiality but fails on time course; Chronos is the DepMap standard for cancer panels but overkill for single-line screens.
For high-stakes hits (drug-target nomination), require 2-of-3 or 3-of-3 method consensus. Single-method hits at FDR 0.05 carry ~5% false discovery; tier-1 consensus shrinks this dramatically.
BAGEL2 BF >6 corresponds to FDR <3% in the Hart 2017 G3 calibration (BF >3 is the FDR <5% threshold); use this when cross-comparing methods.
For low-quality screens (CEGv2 PR-AUC 0.5-0.7), tighten FDR from 0.05 to 0.01 to maintain effective specificity.
Switching from Cas9 to CRISPRi (Dolcetto library) is the cleanest way to bypass copy-number artifact in cancer lines. The tradeoff: CRISPRi knockdown is less complete than Cas9 KO.
For single-cell screens where each cell must carry exactly one sgRNA, target MOI 0.3 (~26% of cells infected, ~4% multi-infected). Higher-MOI designs are valid only when multi-guide cells are filtered or modelled as combinatorial perturbations.
In vivo screens cannot achieve standard 500x coverage with most syngeneic models. Use focused libraries (3,000-15,000 sgRNAs) or CRISPR-StAR temporal activation.

Decision Cheat Sheet

Screen design	Primary method	Secondary check
Two-condition essentiality	MAGeCK RRA	BAGEL2
Time course	MAGeCK MLE	JACKS
Drug vs vehicle	drugZ	MAGeCK MLE
Multi-screen, same library	JACKS	MAGeCK MLE
Cancer-line panel	Chronos	MAGeCK MLE per line
Single-cell	SCEPTRE (see perturb-seq-analysis)	Mixscape pre-filter
Combinatorial / paralog	MAGeCK MLE + GI scoring	See combinatorial-screens
Variant function (BE/PE)	CRISPResso2 + MAGeCK	See base/prime-editing-screens
In vivo	MAGeCK with animal covariate	Per-animal Stouffer meta

QC Checkpoints

Stage	Check	Pass	Action if fail
Counting	Mapping rate	>65-70%	Check adapter / library format
Plasmid	Gini	<0.1	Re-amplify or re-clone
Endpoint	Replicate Pearson on log-counts	>=0.8 (MAGeCK-VISPR floor)	Drop outlier replicate
Biology	CEGv2 PR-AUC	>0.7	Cas9 selection / timepoint / TSS issue
CN	Spearman LFC vs CN	abs(rho) <0.05 post-correction	Apply CRISPRcleanR / Chronos
Depth	Reads per sgRNA	>300	Re-sequence
MOI	Poisson P(>=2)	<5%	Re-infect at lower MOI

Output Files

File	Source	Description
experiment.count.txt	mageck count	Raw count matrix
experiment.countsummary.txt	mageck count	Per-sample QC summary
screen_cleanr_corrected_counts.txt	CRISPRcleanR	CN-corrected counts
essentiality_rra.gene_summary.txt	mageck test	Gene-level RRA scores + FDR
bayes_factor.txt	BAGEL2	Per-gene Bayes Factor
drugz_output.txt	drugZ	Per-direction Z + FDR
jacks_out_gene_JACKS_results.txt	JACKS	Gene effect + sgRNA efficacy
tier_consensus.csv	Pipeline aggregation	Tier-1/2/3 hit assignment
volcano.png, rank.png	Visualization	Standard hit plots

crispr-screens/library-design - Library composition gates downstream analysis
crispr-screens/screen-qc - Six-stage QC with stage-specific thresholds
crispr-screens/mageck-analysis - MAGeCK RRA + MLE detail
crispr-screens/bagel-essentiality - BAGEL2 Bayes factor classifier
crispr-screens/drugz-chemogenomic - drugZ for chemogenomic screens
crispr-screens/jacks-analysis - JACKS for joint multi-screen analysis
crispr-screens/hit-calling - Cross-method decision tree + tier consensus
crispr-screens/copy-number-correction - CRISPRcleanR / Chronos for cancer lines
crispr-screens/batch-correction - Multi-batch covariate design
crispr-screens/crispresso-editing - Editing quantification
crispr-screens/base-editing-analysis - Variant-function BE screens
crispr-screens/prime-editing-screens - PE screens with PRIDICT2
crispr-screens/perturb-seq-analysis - Single-cell screens
crispr-screens/combinatorial-screens - Cas12a paralog screens
crispr-screens/in-vivo-screens - Bottleneck-aware in vivo design
pathway-analysis/go-enrichment - Hit-list functional enrichment
pathway-analysis/gsea - Pre-ranked GSEA