crispr-screen-design

Overview

crispr-screen-design covers pooled CRISPR screens (knockout, CRISPRi, CRISPRa): choosing libraries and modalities, designing the screen for statistical power, executing selection and sequencing, analyzing with MAGeCK-style tools, and validating hits. Pooled screens are powerful but unforgiving — design errors (low coverage, wrong MOI, bad selection) are discovered after weeks of work.

When to use

• Choosing screen modality: knockout (Cas9) vs CRISPRi vs CRISPRa for the question.
• Library selection: genome-wide vs focused, sgRNA count per gene.
• Screen design: cell number, coverage (cells per sgRNA), MOI, replicates.
• Selection schemes: viability, FACS-based, single-cell readouts.
• Analysis: count normalization, hit calling (MAGeCK, BAGEL), QC metrics.
• Hit validation: individual sgRNAs, rescue experiments.

Core concepts

• Modality choice. Knockout (Cas9 cutting): permanent loss-of-function, best for non-essential genes; essential genes drop out (useful for essentiality screens, confounding for others). CRISPRi (dCas9-KRAB): reversible knockdown, tunable, better for essential genes and lncRNAs. CRISPRa (dCas9-VP64/SAM): gain-of-function, for suppressor screens and drug-resistance mechanisms. Match modality to the biology — a KO screen for essential-gene biology is the wrong tool.
• Libraries. Genome-wide (Brunello, GeCKO: ~4-6 sgRNAs/gene, ~75k sgRNAs) vs focused (kinases, epigenetic regulators, custom). More sgRNAs per gene = more power but bigger screens. Use validated libraries; custom libraries need sgRNA efficiency prediction (Rule Set 2, DeepHF) and off-target filtering.
• Coverage. 500-1000x cells per sgRNA at transduction, maintained through the screen (bottlenecks during passaging destroy representation). For a 75k library at 500x: 37.5M cells transduced — plan incubator space accordingly. Coverage is the screen's sample size; skimping here is the commonest fatal error.
• MOI. 0.3-0.5 (Poisson: mostly single integrants). Higher MOI confounds (multiple KOs per cell); verify by antibiotic-resistance titration or fluorescent reporter. Low MOI wastes cells but keeps interpretation clean — accept the waste.
• Selection. Viability/proliferation (simplest; dropout = essential/sensitizing, enrichment = resistance/suppressor); FACS (reporter-based, e.g. surface marker or fluorescent sensor — bins must be wide enough for coverage); single-cell (Perturb-seq: transcriptome per perturbation — rich but expensive). Match selection stringency to keep coverage in the selected population.
• Controls. Non-targeting sgRNAs (negative; ~1000 for good null distribution), essential- gene sgRNAs (positive controls for dropout screens — if they don't drop out, the screen failed), and safe-targeting controls. QC the screen on controls before believing hits.
• Analysis. NGS of integrated sgRNA cassettes → counts → normalization (median ratio, control-based) → hit calling: MAGeCK (RRA), BAGEL (essentiality), or CRISPhieRmix. Check: library representation (Gini index, missing sgRNAs), replicate correlation, control behavior. Hits need FDR control and effect-size ranking.
• Validation. Individual sgRNA knockouts (2-3 per gene, distinct from screen guides), rescue by cDNA re-expression (gold standard for on-target proof), and orthogonal perturbations (CRISPRi vs KO agreement). Screen hits are candidates until validated — expect 30-70% validation rates.

Practical workflow

1. Design. Modality, library, coverage calculation, MOI, replicates (≥3), controls, selection scheme — all on paper first.
2. Pilot. Transduction titration (MOI), Cas9 activity test (essential-gene dropout or reporter), selection condition calibration.
3. Execute. Transduce at scale, select, maintain coverage through passaging (count cells!), harvest with representation preserved.
4. Sequence. PCR of sgRNA cassettes with UMIs/indexing; sequence to adequate depth (≥100 reads/sgRNA average).
5. Analyze. QC (Gini, controls, replicates) → MAGeCK/BAGEL → ranked hits with FDR.
6. Validate. Top hits with individual guides + rescue; dose-response where relevant.
7. Report. Library version, coverage, MOI, selection details, analysis parameters, and validation rates (including failures).

Common pitfalls

• Under-coverage (the screen's silent killer — bottlenecks during passaging).
• MOI too high (multiple perturbations per cell).
• Wrong modality for the question (KO screen for essential-gene function).
• No positive controls (can't tell a failed screen from a negative result).
• Over-stringent selection wiping out library representation.
• Screen hits published without individual validation.
• Off-target effects ignored (no rescue experiments).