MAGeCK Analysis - Usage Guide

Overview

Run MAGeCK on pooled CRISPR screens for genome-wide essentiality, drug-modifier, time-course, multi-cell-line, and combinatorial designs. Covers mageck count from FASTQ, the alpha-RRA test (mageck test) for two-condition comparisons, the maximum-likelihood model (mageck mle) with explicit design matrices for multi-condition / time-course / batch-aware designs, sgRNA-efficiency injection, normalization strategy (median / total / control / spike-in), and downstream visualization with MAGeCKFlute (R) and MAGeCK-VISPR (Python).

Prerequisites

conda install -c bioconda mageck   # not on PyPI                                  # primary CLI
conda install -c bioconda -c conda-forge mageck-vispr   # not on PyPI                            # interactive QC + dashboard
R -e "remotes::install_github('WubingZhang/MAGeCKFlute')"   # removed from Bioconductor at 3.22
conda install -c bioconda bowtie  # MAGeCK uses for off-target check at design time

Required inputs: raw FASTQ (one per sample), sgRNA library CSV ordered as sgRNA id, sequence, gene, and a sample-condition mapping. For MLE designs, additionally a design-matrix file with Samples plus one column per condition (each a 0/1 indicator), and the baseline column always set to 1.

Quick Start

Tell the AI agent what to analyze: - "Run MAGeCK count + test on my drug vs vehicle screen with median normalization" - "Build an MLE design matrix for a time course (Day 0 / Day 7 / Day 14 / Day 21) and run MAGeCK MLE" - "Compare MAGeCK MLE beta scores with JACKS efficacy-adjusted scores on the same data" - "Diagnose why my MLE output has NaN beta scores for half the genes" - "Pick MAGeCK RRA vs MLE vs BAGEL2 vs drugZ vs Chronos for my experimental design" - "Apply batch-aware MLE on a multi-cell-line panel with cell-line and batch as covariates"

Example Prompts

Two-Condition (RRA)

"Run MAGeCK count then test on my screen. Samples Plasmid, Day0, Veh_r1-3, Drug_r1-3. Use --norm-method median first; if essentialome PR-AUC is low, retry with --norm-method control using ntcs.txt. Output gene_summary at FDR <0.05 and LFC magnitude >1 in both directions."

"Generate a volcano plot of my MAGeCK test output highlighting negative-selection hits (depleted) and positive-selection hits (enriched) at FDR <0.05."

Multi-Condition (MLE)

"Build a time-course design matrix with baseline + day7 + day14 + day21 columns. Run mageck mle, output per-condition beta + Wald-FDR. Identify genes with monotonic beta trends across timepoints."

"I have a multi-cell-line screen across 5 cancer lines and 3 batches. Build the MLE design matrix with cell-line indicators + batch indicators + treatment indicator. Recommend Chronos as an alternative if quality / scale warrants."

Drug-Modifier Screen

"Run MAGeCK test on a chemogenomic screen comparing drug vs vehicle (not Day 0). Use --norm-method control with NTCs. Compare to drugZ output: which method is more sensitive for synthetic-lethal interactions?"

"My drug screen has selection at MOI 0.3 then vehicle vs drug 14 days post-infection. Pick MAGeCK MLE with dose covariate or drugZ; explain the tradeoff."

Diagnostics

"Diagnose why my MAGeCK MLE shows NaN beta scores for 40% of genes. Check coverage, sgRNA dropout, and design-matrix singularity."

"My MAGeCK RRA outputs every gene at FDR <0.01 in a heavy-selection screen. Diagnose and recommend a different normalization or hit-calling method."

"Reconcile MAGeCK and BAGEL2 disagreement on my essentiality screen. Which set of hits should I trust for follow-up?"

Visualization and Pathway Analysis

"Run MAGeCKFlute FluteRRA on my output. Generate volcano + rank + square plot + KEGG enrichment for negative selection."

"Launch MAGeCK-VISPR on the screen to generate an interactive QC dashboard."

What the Agent Will Do

Verify library CSV format and sample-to-FASTQ mapping
Run mageck count with sample labels and --trim-5 (a trim length, or AUTO), capturing the per-sample QC summary
Inspect Gini, mapping rate, % zero-count from countsummary.txt; if any fails, refer to screen-qc skill
Decide RRA vs MLE based on experimental design (decision tree in the SKILL)
For RRA: run mageck test with appropriate --treatment-id, --control-id, --norm-method (median default; control if heavy selection)
For MLE: build design matrix from condition assignments; ensure baseline column is all-1; one column per condition or covariate; run mageck mle
For drug screens: vehicle (not Day 0) as control; use NTCs for normalization
Optionally inject sgRNA efficiency from JACKS run on matching cell line
Filter results at FDR <0.05 and LFC threshold appropriate for screen type
Generate volcano + rank + replicate plots; run FluteRRA / FluteMLE for pathway analysis
Reconcile with alternative tools (BAGEL2 for essentiality, drugZ for chemogenomic, Chronos for cancer-line panels)
Output the gene_summary, sgrna_summary, an audit report explaining normalization choice, hit count at FDR thresholds, and method-comparison summary

Tips

Always sequence the plasmid pool and use it as baseline; comparing to Day 0 of the screen pool conflates cloning bottleneck with biology.
For chemogenomic / drug screens, use vehicle (matched DMSO or carrier) as control, not Day 0; drug-induced shifts only mean against vehicle.
Median normalization fails when >40% of guides change. Symptom: every gene significant. Switch to --norm-method control with NTCs.
For multi-cell-line, multi-batch screens, MAGeCK MLE works but Chronos is the DepMap standard with built-in CN bias and screen-quality modeling.
NaN beta scores in MLE indicate a fitting failure (typically <2 non-zero sgRNAs per gene per condition). Exclude these from interpretation, do not treat as zero.
For low-effect-size screens (FDR <0.1 hits at LFC 0.3-0.5), MAGeCK results need orthogonal validation; the statistical model is sensitive enough to call hits that don't replicate.
--variance-estimation-samples lets you fit dispersion from early-timepoint samples (when most guides are still neutral), then test against later samples; useful for low-replicate-number screens.
For paired designs (each replicate from a matched donor), add a paired-replicate column to the MLE design matrix.

Standard Thresholds

Threshold	Value	Rationale
Hit FDR (gene-level)	<0.05	Standard publication threshold
Hit LFC magnitude	>1 (2-fold)	Biologically interpretable effect size
Replicate Pearson on log-counts	>0.85 (acceptable), >0.95 (excellent)	Pre-test QC
sgRNA mapping rate	>65%	Library quality
Gini coefficient (plasmid)	<0.1	See [[screen-qc]]
sgRNAs needed for MLE per gene per condition	≥3 with non-zero counts	Below this, NaN beta
RRA permutation passes per gene	100 default; raise via `--additional-rra-parameters`	Trades runtime

Method Comparison Cheat Sheet

Screen design	Primary method	Backup / sensitivity check
Two-condition essentiality	MAGeCK RRA or BAGEL2	The other
Time course	MAGeCK MLE	JACKS
Multi-cell-line panel	Chronos or MAGeCK MLE	BAGEL2 per-line then meta-analyze
Drug screen	drugZ	MAGeCK MLE with dose covariate
Combinatorial / paired guide	MAGeCK MLE with GI scoring	See [[combinatorial-screens]]
Single-cell screen (Perturb-seq)	SCEPTRE	See [[perturb-seq-analysis]]

crispr-screens/screen-qc - Pre-MAGeCK QC including Gini, replicate, PR-AUC
crispr-screens/library-design - Library design dictates normalization choice
crispr-screens/jacks-analysis - sgRNA-efficacy-aware analysis
crispr-screens/bagel-essentiality - Bayes-factor essentiality alternative
crispr-screens/drugz-chemogenomic - drugZ for chemogenomic / drug-modifier screens
crispr-screens/hit-calling - Cross-method decision tree and reconciliation
crispr-screens/copy-number-correction - CRISPRcleanR / Chronos for cancer-line correction
crispr-screens/batch-correction - Building batch-aware MLE design matrices
pathway-analysis/gsea - Downstream GSEA on MAGeCK ranks