Abstract Optical pooled screening (OPS) has emerged as a powerful technique for functional genomics, enabling researchers to link genetic perturbations with complex cellular morphological phenotypes at scale. However, OPS data analysis presents challenges due to massive datasets, complex multi-modal integration requirements, and the absence of standardized frameworks. Here, we present Brieflow, a computational pipeline for end-to-end analysis of fixed-cell optical pooled screening data. We demonstrate Brieflow’s capabilities through reanalysis of a CRISPR-Cas9 screen encompassing 5072 fitness-conferring genes, processing more than 70 million cells with multiple phenotypic markers. To accelerate biological interpretation, we additionally present MozzareLLM, a framework leveraging large language models to identify biological processes within phenotypic clusters and prioritize gene candidates for experimental validation. Our combined analysis recovers coherent biological modules missed by existing analytical approaches, including five core mitochondrial sub-programs absent from the original study. The modular design and open-source implementation of Brieflow facilitates the integration of new analytical components while ensuring computational reproducibility and improved performance for the use of high-content phenotypic screening in biological discovery.
Bernardo et al. (Sat,) studied this question.