Student Keio University School of Medicine Shinjuku-ku, Tokyo, Japan
Abstract: Perturb-seq is a powerful combinatorial approach that integrates single-cell RNA sequencing with CRISPR-mediated perturbation to profile transcriptome-wide regulatory effects at single-cell resolution. This emerging technology has been widely applied across biomedical research. Perturb-seq typically employs a simple and cost-efficient experimental design, in which genes are knocked down one at a time, to elucidate complex gene regulatory mechanisms in both physiological and pathological contexts. While many Perturb-seq studies rely on this standard design, its statistical properties remain underexplored. Given that the experimental design defines what can be observed—or potentially overlooked—a deeper understanding of its implications is crucial for capturing the full complexity of biological regulation. In this study, we show that even when perturbations are applied to single genes, higher-order interactions can influence the observed outcomes. We also demonstrate that the performance of the standard Perturb-seq design is sensitive to the selection of targets. Moreover, we find that the underlying regulatory network architecture of the biological system critically shapes the effectiveness of the design. By comparing the standard Perturb-seq design to alternative experimental strategies—such as the Plackett–Burman (PB) design—we show, through simulations, that performance varies substantially with different network connectivity patterns. To guide experimental choices, we propose a novel criterion to preliminarily identify whether the standard or PB design is more suitable for a given network structure. These findings offer a conceptual framework for tailoring Perturb-seq protocols to the complexity of biological systems and highlight the importance of design-aware interpretation in high-throughput functional genomics.
