The problem. A pooled screen ends as a table of guide RNA counts before and after selection. Turning that into a trustworthy list of hit genes is harder than it looks: each gene has several guides that disagree, counts are noisy at low depth, and you need genes, not guides, as the final answer. Naive fold-change ranking gives false hits.

The idea. MAGeCK models the guide counts with a negative binomial distribution, the same family used for RNA-seq counts, then normalizes and tests each guide for enrichment or depletion. It aggregates a gene’s guides with a robust rank method, so one noisy guide does not sink or inflate a gene, and reports genes ranked by how consistently their guides moved. It handles both positive-selection and dropout screens.

Why it matters. This is the analysis half of the screen story, and it sits squarely in my own stack. The negative binomial and the multiple-testing care are the exact tools I use for differential expression in aml_rnaseq_nf with DESeq2 and edgeR, reused here for guide counts. Seeing the same statistics carry across assays is the point.

Verdict. The standard analysis method for pooled CRISPR screens, and open access. Read it right after GeCKO to close the loop from design to called genes.