Controlling False Discovery in CRISPR Screens

Abstract

Excluding false positives is critical for interpreting CRISPR screens. Here, we introduce a new Chronos module for estimating false discovery rates for identifying knockouts that cause loss of viability or have differential viability effects in different conditions. We introduce a rigorous benchmarking framework using real CRISPR data. We show with multiple real datasets that existing methods such as MAGeCK are miscalibrated and can generate uncontrolled numbers of false positives even after multiple hypothesis correction. Only Chronos correctly controls false discovery for all tested tasks. Additionally, Chronoss estimates are well-calibrated, allowing users to accurately specify the acceptable false discovery rate.