bio-crispr-screens-hit-calling

Use MAGeCK RRA for standard two-condition essentiality screens; it is fast and robust to outliers. Use MAGeCK MLE for time-course, multi-condition, or drug screens, since it models a design matrix and supports sgRNA efficiency weighting.

drugZ is the primary method for drug-modifier screens because its bidirectional Z-score detects small-effect sensitizers and suppressors that MAGeCK RRA misses. Use MAGeCK MLE with a dose covariate as a confirmatory check.

Merge gene-level results, call MAGeCK hits at FDR below 0.05 and BAGEL2 hits at Bayes factor above 6, then count supporting methods per gene. Genes called by all methods are Tier 1 high-confidence hits requiring minimal validation.

Chronos models copy-number bias and screen quality jointly, while MAGeCK does not, so amplified loci like ERBB2 appear as false MAGeCK hits. Apply copy-number correction before MAGeCK or trust Chronos for cancer-line panels.

BAGEL2's bf step uses a clock-derived random seed by default, so unseeded reruns can flip gene calls at the BF>6 threshold. Always pass a fixed -s seed and verify byte-identical reruns before trusting results.

Avoid RRA when more than 40 percent of guides change, since heavy selection breaks median normalization, and for time-course or cancer-line screens with copy-number bias. Use BAGEL2, MAGeCK with control normalization, or Chronos instead.