Canon enables causal inference of downstream genes in single-cell CRISPR studies via instrumental variable analysis

Abstract

Significance Understanding how genes regulate one another is fundamental to biology, yet existing methods for single-cell CRISPR (sc-CRISPR) screens rely on association-based analyses that cannot establish causality and are prone to high false positive rates. We present Canon, an instrumental variable (IV)-based causal inference framework specifically designed for sc-CRISPR studies, which uses the perturbation status of guide RNAs as IVs to identify causal gene–gene relationships with rigorous type I error control and high statistical power. Applied to real sc-CRISPR datasets, Canon uncovers candidate therapeutic targets with potential relevance for cancer biology and metagenes with distinct biological functions in human lung cancer, demonstrating the transformative potential of sc-CRISPR screening to resolve causal regulatory networks at an unprecedented scale.